THE COMPLETE TECHNICAL PAPER PROCEEDINGS FROM:

ADVANCED MONITORING OF SWITCHED BROADCAST SYSTEMS Ludovic Milin and Ran Oz BigBand Networks

Abstract such as an unanticipated growth, that may lead to insufficient bandwidth being available Switched broadcast has become a viable to support all channel requests. Early technology for reclaiming bandwidth and identification of such trends can allow pro- optimizing spectrum, allowing richer program active remedies, minimizing negative lineups and increasing personalization of subscriber impact. content. By leveraging switched broadcast The authors also highlight the imperative cable operators are able to offer subscribers of protecting consumers’ privacy and hundreds of high definition programs and describe how the techniques used to collect long tail content, such as coverage of local viewership data can be made to comply with sporting events, without requiring plant the 1984 US Cable Privacy Act. upgrades to expand capacity. This paper presents data gathered from INTRODUCTION multiple switched broadcast deployments to illustrate the imperative of monitoring system Switched broadcast came of age recently performance and the resulting benefits to as two major North American cable operators operators. deployed the technology in several markets. The authors assert that performance Switched broadcast currently supports over monitoring should be applied at three key six million homes passed and switches one stages of deployment of a switched broadcast million set-top boxes. With other cable system: operators evaluating the technology and undertaking field trials, the footprint of 1) Prior to Deployment – during this stage deployed switched broadcast systems is set to an operator must assess which programs in its grow rapidly. broadcast lineup are viable candidates for putting on a switched tier. A non-intrusive By reclaiming the bandwidth that would analytical tool is, arguably, the best way to otherwise be consumed by delivering collect viewership statistics for all available unwatched content, switched broadcast programming, allowing identification of provides cable companies the opportunity to “long tail” content. expand the amount of programming they offer subscribers. However, the oppprtunity to 2) During Deployment – while a switched provide thousands, even tens of thousands, of broadcast system is being implemented an programs requires operators to evolve the operator will benefit from leveraging the methodologies used to track viewership of same monitoring platform used in the prior that content. The ability to collect statistics, in characterization effort, potentially still real-time, about which programs subcribers gathering viewership data on broadcast are watching is necessary for successful programming. capacity planning and deployment of switched 3) After Deployment – once the installation broadcast systems. The insights gained from has been completed the monitoring focus can collection of this data facilitates rapid and be broadened to allow identification of trends, smoother deployment of switched systems. Valuable statistics include the viewership • Diagnostics; of linear and switched broadcast programs, • Content personalization; the number of active STBs (set-top boxes) per service group, and the bandwidth utilization • Addressable advertising; within service groups. • Market research. Additionally, the ongoing collection of This paper discusses each of these information enables round-the-clock applications in depth and cites the specific performance monitoring, leading to benefits of a statistics monitoring platform in opportunites for pro-active network each instance. It begins, however, with a maintenance. Data about blocking events, description of how the platform functions. failures for STBs to tune to the requested channel and network response times can all be ARCHITECTING NETWORKS FOR analyzed, leading to enhanced performance STATISTICS RETRIEVAL and, ultimately, increased subscriber satisfaction. The statistics monitoring / diagnostics In addition to the operations benefits platform proposed by the authors consists of obtained from deploying a statistics the following components: monitoring tool, the ability to build accurate • Collection Engine – gathers STB activity and complete records of viewership behavior logs from the servers, known as SBSSs introduces new revenue possibilities. (switched broadcast session servers), Personalized news is one example of ways used to manage a switched broadcast that precise viewership records collected by a network; statistics monitoring tool could be leveraged for new revenue generation. Another is • Analytical Engine – processes the STB addressable advertising, which matches ads activity logs gathered from the network; more closely to subscribers’ interests, by • Data Warehouse – stores the data and leveraging insights into which content metadata obtained from the both the viewers are habitually tuning to. network and analytical components, and Finally, while the use of third-party market supports long-term storage for trending research firms has generally been sufficient to analysis; obtain insights into subscriber viewing habits, the plethora of new programs that will be • Webserver – provides a user-friendly available to subscribers in switched interface to customize and view reports. environments makes this task more daunting. Figure 1 shows how these elements are The value of this information is high because combined in a generic switched network. it provides insights into the viewing patterns of all subscribers on the switched tier, not just the subset of viewers that have been enlisted and whose viewing habits may not necessarily represent those of the majority.

This paper describes a method for collecting statistics that supports the applications described above, namely:

• Capacity planning;

Figure 1: Components of a statistics management platform

Rather than the traditional store and The real-time metrics derived from the forward approach in which each STB protocol messages can be grouped into dedicates a portion of its mem ory footprint to general audience metrics and specific recording activity and, on a regular basis, switched broadcast performance metrics. sends large UDP (user datagram protocol) packets over the OOB (out-of-band) upstream General audience metrics (for both path to some collection server, the authors switched and linear programming) include: recommend an approach that has no impact on • Number of viewed channels per the STB memory footprint or the OOB service group; network. This feature stems from each message requiring only about 64 bytes. This • Overall bandwidth utilization per approach, which leverages standard switched service group; broadcast protocols, is not sensitive to packet • Number of active STBs per channel loss between an STB and the data collection and per service group; engine. It is not tied to a specific switched broadcast client or headend environment – • Number of tune-ins and tune-outs per any LOB (load on boot) or native switched channel and per service group. broadcast client can be utilized to send STB Switched broadcast performance metrics activity messages to the SBSS. include: An example of these STB messages is • Switched broadcast QAMs occupation presented in Figure 2. ratio per service group; The same protocols can also be leveraged • Number of tune-ins to previously to monitor linear broadcast channel activity unmapped channels per service group; via simple reconfiguration of the switched broadcast client on an STB. This enables • Time to map previously unwatched activity messages for all channels to be sent channels per service group; upstream instead of only those on the • Upstream bit rate per servce group; switched tier. • Number of blocking events (bandwidth not available) per channel and per service group; • Number of other errors per service group (unresponsive STBs, tuning errors, and so on).

SBSS STB Init Request Init Init Response

Program Select Channel Change Program Response

Query Request STB <-> SBSS Session Maint. Query Response

Event Indication Force Barker / Record Indication Event Response

Force Channel Program Indication Prog. Ind. Response

User Activity User Activity Report Tracking

Figure 2: Example of protocol messages communicated between the STB and the SBSS

The following assumptions were made and enables the analysis it performs to be during development of the statistics collection more effective / broadly applicable. and analysis tool described in this paper: The figures presented throughout this • A service group is a group of nodes; paper are screen captures from the web-based user interface of the diagnostics tool. The • Each node serves multiple households reports can also be configured to output pdf • A hub serves multiple service groups; files for easy sharing amongst a working group, and CSV (comma–separated values) • An SBSS supports multiple service files so the raw data in the reports can be groups; analyzed further. The data was gathered from • All nodes in a service grup are served deployed switched broadcast systems in by common switched RF spectrum; multiple cable networks.

• The SBSS shares the RF spectrum; APPLYING PERFORMANCE • The SBSS logs all STB activity in MONITORING: THREE SCENARIOS real-time (channel changes, keep alives, last user activity, and so on.) There are at least three scenarios where the A common set of assumptions simplified ability to collect viewership statistics and and accelerated creation of a diagnostics tool performance diagnostics can be especially performance a cable perator should assess valuable: which programs to put on the switched tier. A non-intrusive analytical platform is, arguably, • Before deployment; the best way to collect viewership statistics • During deployment; for all available programming, and enable identification of “long tail” content. This type • After depoloyment. of platform can also provide the parameters Each of these scenarios is discussed in the necessary for modeling switched following sections. infrastructures, allowing for optimization of Edge QAM numbers, service group size and BEFORE DEPLOYMENT over-subscription ratios.

To ensure best-of-breed network

Figure 3: Typical weekly busy hour report (total number of viewed channels for all service groups) Figure 3 provides the total number of population is different, and also to assess the viewed channels (active streams) for all impact of service group size. Examples are service groups for a week. Notice the daily shown in Figure 4 of the maximum number of spikes between 10PM and midnight and more active streams, varying from single to triple at consistent viewership throughout weekend. It constant service group size, indicating that the is possible to generate similar reports for each number of switched broadcast QAMs per service group to check whether the service group can be optimized, instead of conclusions made for the overall population being a constant. apply, to compare service groups that are geographically distributed or whose

Figure 4: Example of weekly busy hour report for an individual service group

Once the busy hour has been identified one accomplished. should analyze the viewership of individual Figure 5 shows a typical report for all channels audience to allow “long tail” content service groups aggregated. Because this candidates suitable for being placed on the example provides insights into channels that switched tier to be identified. It is im portant are already on the switched broadcast lineup to narrow down this analysis to the busy hour “long tail” programs can be identified easily as this is where the most significant bandwidth saving benefits can be

Figure 5: Distribution of “long tail” content

over-subscription ratio. This is achieved by The channel audience in service group can examining the channel concentration ratio, also be used to predict the switched broadcast (i.e. the ratio of unique viewed channels to the total number of available channels in the Using that data, one can again optimize the SWB tier.) Figure 6 shows that this ratio is at number of switched QAMs assigned per most 40%, leading to a minimum 2.5:1 service group. oversubscription ratio in the switched tier.

Figure 6: Channel concentration report

The reports presented in this section As new service groups are being migrated should not only be used in a preliminary to the switched tier one should confirm that switched broadcast deployment study but can STBs are registering with the SBSS and there also be used throughout the life of a switched are no upstream communication issues. An broadcast deployment to ensure that the example of such a report, showing the total conditions are not changing and that the number of active STBs as a function of time, switched infrastructure remains optimized. is provided in figure 7.

DURING DEPLOYMENT A side benefit of this type of report is that it can be used to verify that the number of active A cable operator can benefit from tuners in each service group doesn’t exceed leveraging the same monitoring platform used expectations. Trending growth over time will in the prior characterization effort when indicate when a service group split may be installing / turning-up a switched broadcast appropiate. system: the statistics collected previously can facilitate easier troubleshooting of deployment issues such as STB upstream communication issues and QAM mis- configurations.

Figure 7: Number of active STBs in a specifc service group

Once a channel has been mapped to a 99.9% of the channel changes (not including QAM group a STB will tune to it without the STB decoder actually locking into the delay, a direct benefit of the protocols used in stream) occurred in 5msec or less. switched broadcast. Figure 8 shows that over

Figure 8: Channel change times already mapped on the switched tier

A departure from a 0 to 5msec range already mapped to one of the switched QAMs would indicate an issue with the system and because no one in that service group was along with likely impact to the subscriber’s previously watching it, additional steps will viewing experience. be required for the STB to tune to it. Figure 9 shows a typical distribution of such delays, On the other hand, when a channel is not which include the time for the request to go negligible amount of time in terms of from the STB to the SBSS, the SBSS subscriber experience. This metric can be processing of the request and forwarding to affected by delays in the network or SBSS the Edge QAM, the Edge QAM joining the overloads and is, therefore, important to multicast group for that stream, adding the closely monitor during and, after deployment. session to one of the service group QAMs, Doing so will help address potential issues and updating of the MCP to send the tuning and avoid subscriber complaints about tuning information to the STB. It typically takes delays. under 120msec to accomplish this, a

Figure 9: Additional delay for channel changes requiring mapping of an unwatched switched program

Because the OOB upstream bandwidth is Typical loads vary between 0 and 200Kbps limited it is important to keep track of how depending on the number of active STBs per much is being used by switched traffic flows service group, as figure 10 illustrates. between the STBs and their respective SBSSs.

Figure 10: Maximum upstream bit rate for switched broadcast traffic per service group

The SBSS performance (see figures 8 and of an SBSS to split the load and maintain an 9) is a strong function of the number of acceptable level of service. channel change requests the server has to Because the viewership and performance handle. That metric is presented in Figure monitoring tool provides real-time visibility 11. Once a baseline has been established one into the system behaviour, there is no need for can monitor how this metric changes with the empirical rules, since action can be taken as number of active STBs (see figure 7) and soon as the measured subscriber experience measure the resulting impact on SBSS threatens to dip below the deemed acceptable performance. A typical scenario would be an thresholds. increase in the number of STBs and their respective traffic flows, requiring the addition

Figure 11: Maxmum number of channel change requests per second per service group

The number of channel change requests high. As can be seen from the data provided requiring a new program t be mapped to the in figure 11 and 12, the ratio between the two service group lineup should typically be fairly is about 1000 to 1. This ratio, however, does small unless the switched broadcast channel change significantly from one system to map is large and the over-subscription ratio another. A large and sudden departure from the established baseline would typically the MCP reaching the STBs. indicate issues with system configuration or

Figure 12: Maxmum number of channel change switch requests per second per service group

AFTER DEPLOYMENT available within each service group and the number of active streams (actively watched After installing a switched broadcast channels). It also encompasses the bandwidth system the monitoring focus can be broadened of individual streams, something that is not to allow identification of trends such as un- necessarily constant when using multi-width forecasted growth that may lead to CBR (constant bit rate) to maximize video insufficient bandwidth being available to quality, or when mixing standard and high support all requested channels or management definition progams. Trending this metric over servers becoming overloaded. Early time for each service group will characterize identification of such trends can allow pro- the growth rate, and a good rule of thumb is to active remedies, minimizing negative consider adding more QAMs or splitting subscriber impact. service groups when the ratio consistently surpasses 80 to 90%, or spikes to 100%. Experience has shown that one of the most relevant metrics to follow is the occupied bit Figure 13 shows one such daily summary of rate concentration ratio. This metric takes into the maximum occupied bitrate concentration account the number of switched QAMs ratio for each service group.

Figure 13: Maximum occupied bit rate concentration ratio per service group

Figure 14 examines the top three service during the typical 10PM to midnight busy groups in Figure 13 in more detail. One of the hour, meaning that there is no immediate need key things to note here is that the 80% for any infrastructure changes. utilization threshold is only surpassed briefly

Figure 14: Daily variation of occupied bit rate concentration ratio for three service groups

Once the occupation ratio is consistently in is greater than 1:1, the likelihood is high that 100% vicinity and the over-subscription ratio no bandwidth will be available on the QAMs for new channels. This is defined as a each service group, of the number of blocking blocking event and results in a subscriber events over the duration of a few days. receving a “Channel not available – Please try Approximately 15% of the service groups later” message, instead of the desired show a significant number of blocking events. program. Figure 15 presents a summary, for

Figure 15: Maximum number of blocking events for all service groups

As a remedial step to bandwidth scarcity the service groups under consideration, the additional switched QAMs can be added to number of active streams on the switched tier the service groups most affected by blocking. jumped from about 20 to about 30 as soon as The results of doing so ae highlighted in the additional QAM became available. Figure 16. This example shows that, in one of

Figure 16: Maximum number of viewed channels for a specific service group across the addition of a switched QAM Coincident with the increase in available Additionally, the over-subscription ratio bandwidth the numbers of blocking events decreased from 2.5 to 1.7 in the affected virtually disappeared, as figure 17 shows. service groups

Figure 17: Blocking events for a specific service group across an additional switched QAM

This can be explained by the larger size of introduces new revenues opportunities. These the service groups, subscriber demography are examined in the following sections of this and switched channels lineup. The positive paper. point is that because of the detailed monitoring information available, the capital CONTENT PERSONALIZATION needed for an additional switched QAM was limited to a very small number of clearly Personalized news is one example of ways identified service groups. Moreover, the that precise viewership records collected by a expense could be delayed until the blocking statistics monitoring tool could be leveraged situation became too affecting for the for new revenue generation. Since the real- subscribers. time collection of viewership data allows content and subscriber interests to be BUSINESS BENEFITS OF STATISTICS accurately correlated, enterprising newrooms, MONITORING or third-parties, could create news summaries that are more likely to retain the attention of Benefits accrue when statistics monitoring viewers, as compared to traditional broadcast is applied to both linear and switched TV news programs. For example, newsrooms broadcast networks. In addition to the could record a series of short news stories on operations benefits already described, namely a wide range of topics, allowing cable capacity planning and diagnostics, the ability operators to collect accurate viewership data full-time to combine into personalized bulletins that how three subscribers, all watching the same address a subscriber’s specific interests, program on the switched tier receive different whether it’s baseball teams or Internet start- ads during the commercial breaks. For ups. example, subscriber #1, an avid teen snowboarder, receives an ad about snowboard A personalized version of a music network sales. During the same commercial break, is another example of how switched unicast subscriber #2, a thirties-something bachelor, provides cable operators the opportunity to views an ad about an upcoming motor show. offer increasingly customized content. Subscriber #3, an avid traveler in her fifties, Customized services like these could be receives information about cruises in the offered as premium services, potentially Caribbean. earning the operator additional revenues. Studies, such as a recent CTAM report, Alternatively, personalized content could be reveal that marketers are willing to pay offered at no additional charge, with the premium rates if their ads achieve improved expectation that increased customer loyalty response rates among their intended results. audiences.

ADDRESSABLE ADVERTISING Addressable advertising is a methodology By building precise databases about which programs individual subscribers are watching, for more closely matching advertisements to subscribers’ interests. Figure 18 illustrates

Clamping Platform Switching Subscriber Platform SWB SG RF #1 RF Combiner Diplexer Subscriber #2 Mgmt Server Subscriber #3 OOB Mod

Ad OOB Demod Server

Figure 18: Subscribers receiving different ads though all view the same programming cable operators can determine which ads to candidate for an advertisment about a relevant forward to viewers. local fashon event. Other methods of recording consumer viewing habits are In contrast to third-party research firms unlikely to yield this level of detail. that typically collect broadbrush viewership data, the statistics monitoring platform Additonally, a consumer that routinely described in this paper can provide insights tunes to a home improvement network may be that would otherwise be lost. For example, a receptive to an advertisement about a sale at a subscriber, that may have paused to watch a local hardware store, and could receive such short segment about the fashion industry an ad, even when watching a different while channel surfing, may be a viable network.

GUARDING PRIVACY RIGHTS CONCLUSIONS

The Cable TV Privacy Act of 1984, is The ablity to obtain precise viewership intended to protect personal information. In statistics assists cable operators in a variety of particular, it prohibits cable TV providers ways. These include providing the insights from disclosing personally identifiable needed to engineer effective switched tiers information, and allows users to view and and speed system deployment and turn-up. verify their information. Naturally this means The capability to monitor network that gathering and reporting on an perfomance allows adjustments to be individual’s channel viewership history is implemented that ensure channel change likely illegal. In order to satisfy this times and service quality are optimized. requirement, the SBSS servers offer the Although primarily developed for switched option to scramble the STB MAC addresses in broadcast environments, the non-invasive tool the logs making it impossible to identify any described in this paper has applications in single user. linear broadcast networks also. Broader The gathering of individual subscriber data applications can include support for can be implemented on an “opt-in” basis. For addressable advertising business models and example, the more enterprising cable other content personalization oppportunies. operators could explicitly ask their However, maintaining subscribers’ privacy subscribers about their ad preferences. In is paramount and the viewership statistics return for providing a cable operator with a needed to empower these benefits can be list of the subjects and categories they’d be obtained / stored in ways that enable cable interested in viewing ads on, subscribers operators to meet their obligations under the could receive a complimentary gift or upgrade Cable Privacy Act. to an expanded service package, or some other incentive.

CABLE THIN CLIENTS: A NEW REVENUE OPPORTUNITY FOR MSOS Daniel Howard Motorola Connected Home Solutions Technology Office

Abstract you plug a monitor, keyboard, and mouse into, MSOs can enter the market at far less than A thin client computer typically a PC that con- would be required for conventional computer sists of a motherboard only and boots the op- technology. Finally, since the servers are lo- erating system off of the network. Thin clients cated in the MSO's facilities, the operating are essentially intelligent display devices that system is completely under the MSO's control, normally run applications on a server located and thus can be easily maintained and provide somewhere else on the network. The lack of an additional portal to users. moving parts and ability to manage them re- motely makes them a very cost effective tech- nology for business applications such as point INTRODUCTION of sale devices, and many corporations are using thin clients for their employees based on With the growth of competition for cable op- the much lower total cost of ownership of this erators in the triple and quadruple play arenas, technology. Recently, with the availability of there is renewed interest in emphasizing the Open Source office applications and educa- content and services provided by the operator tional titles, schools all over the world are to subscribers. And this content is not just switching to thin clients as a more cost effec- limited to TV, music, and Internet portals, but tive approach to providing reliable and inex- many operators are providing free software, pensive student PCs. such as anti-virus applications, along with their high speed data services. These software In this paper, a thin client computer technol- application content offerings are generally be- ogy that uses cable modem infrastructure is ing used today to enhance the marketability of described. Because of its broadcast architec- high speed data service, rather than to specifi- ture, the cable plant is uniquely positioned to cally generate new revenue streams. take advantage of thin client computing, and can provide the service more efficiently and The main drawbacks of offering software con- cost effectively than other broadband wired tent are that it can increase the software sup- networks. Thin clients are as robust as a cable port cable operators must provide to subscrib- modem, and can be rebooted and managed ers, and the multiplicity of platforms and oper- remotely as easily as a cable modem, which ating systems to support can make such con- means MSOs can offer a managed home PC tent offerings undesirable to the operator. The service without concern for additional truck cost of licensing such software can also be a rolls. Best of all, it opens up new revenue op- burden to operators’ revenue growth. portunities for MSOs, including pay-as-you-go data services, children's PCs with preloaded Businesses have been plagued by these same educational titles, and elderly residents who issues in providing PC platforms and software would only get a PC if a service provider sup- applications to their employees since the ad- ported it for them. And since a cable thin cli- vent of the PC. An approach that is currently ent is not much more than a cable modem that gaining ground in the business community is to use a thin client architecture to reduce the the cable operator can standardize both the total cost of ownership and reliability of em- hardware and software for the system, and can ployee PCs. Thin clients are essentially a PC manage the clients from the operator’s facili- that consists of a motherboard only, and boots ties, much as cable modems and settop boxes the operating system off of a server located are managed currently. For the purpose of this elsewhere in the network. In a typical thin cli- paper, this proposed PC platform is termed a ent deployment, over one hundred clients can cable thin client. be run by a single server, which means the business IT department need only support a An introduction to thin client technology, an single PC, the server, and the thin clients are architecture for, and the challenges of deliver- more like appliances that at most require an ing thin client computing over cable networks occasional rebooting. Thin clients are much are presented in this paper. Benefits to both cheaper than conventional desktop PCs ($150 consumers and operators are provided, as are vs. $300+), and software additions, updates, new MSO revenue opportunities. Finally, fu- patches, etc. are done on a single server, and ture network architectures that would include then apply to hundreds of clients. cable thin clients are referenced, and a sum- mary of why cable thin clients make sense for To further reduce the total cost of ownership cable operators is presented. (TCO), many businesses and schools are switching to Free Open Source Software THIN CLIENT TECHNOLOGY (FOSS), and indeed an increasing number of cable operators are requesting Open Source While most PCs can be converted into thin cli- solutions for their networks and services. As ents merely by altering the BIOS to boot the opposed to proprietary software offerings, PC off of the network instead of the hard disk FOSS is developed by the world community drive, the real benefits of thin clients can be and the source code is freely available to pro- seen in platforms designed to operate only as a grammers to download and improve. The thin client. Most thin clients are about the size Linux operating system is probably the most of a cable modem, but there are also thin client famous FOSS package, but FOSS applications solutions built into LCD monitors and even exist for all major operating systems, and in wall data ports, such as the Jack-PC from Jade fact there is now an Open Source office suite, Integration as shown below: OpenOffice, that rivals proprietary offerings in (www.jadeintegration.com/jackpc.php). function, power, and compatibility.

So why would a cable operator be interested in thin client computing and Open Source Soft- ware? Because these are enabling technolo- gies for cable operators to offer new software content and new high speed data services to subscribers. An example new service that would generate an entirely new revenue stream Figure 1. Example Thin Clients. for operators would be to offer a home PC platform, running Open Source Software, that Most are diskless and fanless, which means the can be cost-effectively deployed and that is moving parts that typically fail in a home PC easily supported by the operator. By using are completely eliminated, and so thin clients thin client technology as the home platform, are projected for TCO purposes to have 7-10 year lifespans, or roughly twice that of con- uted to as many servers as desired, and indeed ventional PCs. During its life, all thin clients install disks could be sent home with students in the network may be upgraded by upgrading so they could use the software at home as well the servers. as at school. The package used, K12LTSP, includes dozens of educational applications, The basic thin client architecture is to have at with an emphasis on math and science. After least one server on the network from which the the initial deployment, and with over 100 do- thin clients boot their operating system and run nated PCs from local businesses, a seachange applications, as shown below. occurred in how the PCs were used in the classrooms, and test scores even went up dra- Local or matically the number of working PCs in each room was tripled. The result is that the public Server Wide Area school district of Atlanta is now deploying this Network technology, using brand new thin clients the switch size of a cable modem, servers which support over 100 clients each and LCD monitors. The LCD monitors were chosen because when Thin Thin Thin classroom ratios of student to PCs of 2:1 are approached, the electricity available in a room Client Client Client becomes the limiting constraint, and lower Figure 2. Basic Thin Client Architecture. power-consuming devices must be used. A a new thin client with an LCD monitor uses Thus, the server has two Ethernet interfaces, about 1/6 the electricity of a conventional one to the outside world and one to the thin desktop PC and CRT monitor. clients. The server can either be a single server that provides all required functions of The low cost of thin client technology, the boot, authentication, and applications, or these depth of FOSS software available, and the functions can be divided into servers opti- complete success of this approach for school mized for each function, as is done in enter- districts that typically struggle to maintain prise-style deployments of thin clients. their information technology, led to the pro- posed cable thin client solution provided in The author’s own initial experience with thin this paper. client technology using Free Open Source Software was at his child’s elementary school ADAPTING THIN CLIENT TECHNOLOGY several years ago, where classroom computers TO CABLE DATA NETWORKS were 7-10 years old and typically unused by teachers and students due to lack of adequate To implement thin client technology over a maintenance. After installing classroom serv- cable data network, several modifications to ers in each room, the three existing PCs were the basic architecture must be made. First, the suddenly brought back to life, working faster enterprise version must be implemented, than ever before, and maintenance issues all wherein powerful servers in the cable opera- but disappeared, even for PCs that were up to tor’s facilities are used. This is so that truck 10 years old. Furthermore, the software suite rolls likely involve using separate servers for used was completely sufficient for all of the boot, authentication, file, and application serv- teachers’ and students’ needs, and the fact that ing. There are two approaches currently used it was free meant that copies could be distrib- for application servers: first, use multiple ap- plication servers that each have all of the ap- architecture, if more than 6 or 7 clients are plication software. This approach has the connected to a server, then Gigabit Ethernet is benefit of providing load balancing and auto- required for the link between the switch and matic failover if a server has issues. The sec- the server, which aggregates the traffic of all ond approach is to use application servers that clients. However, most applications, like have been optimized for specific applications. OpenOffice and even video streaming from This approach often optimizes performance in websites like YouTube and CNN took only 1 the system, but unless it is combined with re- Mbps or so per client. dundancy, can lead to unhappy subscribers. Nevertheless, there are already available com- A proposed architecture for a cable thin client pression packages such as the FreeNx package (TC) network is shown below. Note that four (http://www.nomachine.com), which can be types of CPE are shown: standalone ca- combined with Linux terminal server pack- blemodem (CM) and standalone TC; com- ages, and purports to offer thin client support, bined CM/TC, and then two similar variants even over dialup networks. Experiments with where the CPE also includes a server. These this package indicate that while the latency is latter two approaches would be applicable to noticeable, it is tolerable, and clearly an opti- MDUs and small businesses.are not required mization of such an approach would lead to to manage and maintain the servers, as well as better user experience, while keeping the providing additional security for the servers bandwidth requirements low. and control over the user interface presented to clients. This would also But cable data networks can take advantage of their broadcast architecture to enact further Customer improvements in delivering thin client screen Application Premises Client ServersApplication Boot Thin ServersApplication Server(s) Client data. The desktop itself can be standardized ServersApplication Client Servers Customer Thin Application Client Client Servers Premises such that much of the screen data is common Authentication/ File Thin Local Server(s) Client Server across many thin clients. Further, as an over-

CM CM load protection for events such as popular web Headend WAN Switch or Cable Modem HFC Backbone Termination System video suddenly becoming available, the video Transport (CMTS) Network Adapter Combined CM/ streams can be delayed and synched so that Local Server Security and Combined CM/Thin Backbone Access Thin multicast delivery efficiencies can be obtained. Client Network Controller Client Local Thin Also, the thin client platform can be made Server Customer Client Facility Premises Customer thicker so that certain especially bandwidth Premises Remote Server intensive applications can be run locally. Fi- WAN Facility nally, an adaptive system, which trades off cli- ent bandwidth requirements vs. latency can be Figure 3. Example Cable Thin Client Net- used. work. BENEFITS FOR THE SUBSCRIBER The bandwidth requirements of thin client net- works can be substantial. Without any form of Why would a customer pay a monthly sub- compression of the screen data before sending scription to have a headache-free PC in the to the thin clients over the network, a single home loaded with software that is maintained thin client can require several Mbps down- by the cable operator and replaced by the op- stream data rate, and for particularly animated erator if it ever fails? First, there are many applications, a client can require over 10 Mbps subscribers, especially the low income and the downsteam data rate. Hence, even in the basic elderly, for whom the desktop PC is either too Interestingly, one other user benefit of this ar- expensive to purchase initially, too compli- chitecture is that a user’s experience is not cated to decide what to get, or too worrisome fixed in space: a user can go to a neighbor’s to have in the home unless a service provider home, Internet café, etc., and can log onto the supported it. Second, many homes with chil- cable modem network there, perhaps requiring dren still do not typically have a PC dedicated an encrypted USB key, and then their desktop to the children’s use, which means the children environment, including all services to which end up using the parent’s PCs and often the they currently subscribe, are available. This web sites for kids end up downloading soft- kind of seamless mobility is a new wrinkle on ware that slows the PC down. Were the chil- PC mobility. In the school environment de- dren’s PC to be a cable thin client however, scribed earlier, this architecture has proven to since the server is in the cable operator’s fa- be a method of providing ubiquitous PC access cilities, the operator can use web content filter- that is less costly and more reliable than 1:1 ing and URL-blocking applications like laptop initiatives, for example. DansGuardian and Squidgard, to provide a safe web accessto all children’s thin clients on CABLE OPERATOR OPPORTUNITIES the network. Further, the operator could in- clude other content related to the cable opera- The main new opportunity for cable operators tor’s business, such as flash-based children’s is a completely new source of revenue from programming, and built in links to cable opera- offering a cable thin client subscription ser- tor’ web sites for children. vice. This addresses both existing high speed data customers as well as a currently untapped The protection of user data and privacy may be base of subscribers. Minimum 2 year con- effected in several ways. First, the user can tracts with free thin clients are a possibility, take responsibility for their own data protec- similar to the cell phone model. And a cable tion and use an encrypted USB flash drive thin client goes beyond even the quad play plugged into the thin client. Second, the cable arena into the so-called x-play scenarios, operator can maintain files in their facility, where new devices leverage multiple services, keeping them private via built-in mechanisms applications, and user experiences; a cable thin in Linux, or using other applications which client is a platform for PC-TV as well as IPTV increase file security. Especially for non- and interactive TV. Since the thin client is a computer literate subscribers, the knowledge “12 inch” interface, but can easily include that files are automatically backed up by the more typical video content, the operator now operator could either enhance the service or has a platform for ITV that moves into the provide additional monthly revenue. Printing, next generation of applications without con- external CD ROM drives and other USB de- cern for the often limited processing power vices can also be installed in the client from and capabilities of older settop boxes. the server, although a support call would typi- cally be required. The opportunity for ad revenue enhancement is also significant. With a cable thin client us- User benefits thus include: all typical office ing a common portal and linked to the opera- applications (web browsing, office applica- tor’s content, the cable operator has unprece- tions, email); children’s edutainment applica- dented influence on the web and video experi- tions; automatic, secure network backup of ence of the subscriber. And by integrating user files; automatic software updates; and all data from such a service with other home ser- at no licensing cost. vices provided by the cable operator, the op- erator has valuable usage data across a variety For the service provider: of platforms and services that can raise ad revenue rates for the operator. • New revenue opportunities

FUTURE DIRECTIONS AND SUMMARY • Better metrics on customers, enhanced content value Looking to the future, when high definition video, ultra high bandwidth data services, • Leverages existing broadcast infra- peer-to-peer video sharing and conferencing structure are all being delivered via cable operator’s networks, the cable thin client represents yet • Leverages existing Open Source appli- another bandwidth consuming service, but one cations that easily integrates with existing services. In these scenarios, other means of delivering high For the consumer: definition video, which can include thin client screen data, may be required. One example is • Maintenance-free computing, auto- the DOCSIS IPTV Bypass Architecture (M. matic updating of software and features Patrick and G. Joyce, “Delivering Economical IP Video over DOCSIS® by Bypassing the M- • Low cost entry to Internet computing CMTS with DIBA, SCTE 2007) as shown be- low, where the thin client servers have been • Security, privacy, and reliability moved deeper into the back office and the thin client uses DOCSIS 3.0 technology to further increase the bandwidth available to the client.

The benefits of thin client computing over ca- ble networks can be summarized as follows:

Client Boot ApplicationClient Server(s) ApplicationServers Servers Customer

Client Premises Authentication/ File Thin IPTV Server(s) Client

DOCSIS 3.0 MPEG/DEPI CM EQAM WAN Converged HFC Interconnect Network Network (CIN) Cable Modem Termination System (CMTS)

Backbone Security Network and Local Access Server Controller Facility

Remote Server WAN Facility

Figure 4. Thin Client Architecture Using DIBA.

CREATING LOGICAL CHANNELS AND IMPLEMENTING ADVANCED SPECTRUM MANAGEMENT Jack Moran and Michael Cooper Motorola Connected Home Solutions

Abstract maximum data signaling rate for logical

channels. ® DOCSIS 2.0 support for logical channels enables cable operators to increase the available upstream bandwidth. This paper They can leverage logical channels and will highlight the business and service advanced spectrum management to increase advantages of creating logical channels and the aggregate bandwidth. Automation is key implementing advanced spectrum to the successful implementation of logical management to monitor performance and channels and advanced spectrum efficiently utilize HFC bandwidth. management.

INTRODUCTION The CMTS, Advanced Spectrum Management and DOCSIS 2.0 Logical Operators can quickly deliver increased Channel Operation can begin the tedious upstream performance to HFC networks process via a script and automatically compile consisting of mixed mode DOCSIS 1.x and the various CMs or MTAs into the up to 4 2.0 Cable Modems (CMs) as well as for new Logical Channel Configurations supported. networks consisting entirely of DOCSIS 2.0 From the perspective of maximum throughput cable modems. But there are issues to address being maintained, it is quite possible to to reap these rewards, and this paper will eventually allow the CMTS to automatically explore them. make the decision regarding Logical Channel Assignments and to affect the change, thus This paper will explain the technology, allowing the cable operator to adaptively discuss the challenges involved in change the maximum throughput possible implementing logical channels, and provide based solely on return path conditions examples of performance gains realized by presented to the CMTS. deploying logical channels in both mixed- mode and pure DOCSIS 2.0 environments. Today, this is not performed due to the cable operator having to gain confidence in It will also discuss how cable operators the Logical Channel estimation process which can utilize a spare receiver on a CMTS to is being performed by doing the analysis perform spectrum management and non- automatically and assigning all DOCSIS 2.0 obtrusively measure spectrum impairments, CMs and MTAs the correct Logical Channel predict the dominant impairment in any area Assignment, but first allowing the cable of the spectrum by analyzing signal-to-noise operator to make the final decision as to the measurements, and qualify any area of the throughput change. Once the cable operator return path spectrum so they can achieve the gains confidence in the accuracy of the throughput estimation via Logical Channel Operation, it is logical to assume that the cable operator will take the next step in the Prior to the introduction of logical process and allow fully automated Logical channels in DOCSIS 2.0, each physical Channel Operation. channel was required to utilize only one value for each of the burst profile attributes. Thus, the configuration of an upstream channel in UNDERSTANDING LOGICAL DOCSIS 1.1 or 1.0 was driven by the worst CHANNELS perform ing CM or tap in the plant. For The concept of logical channels was example, if a CM was located at a point in which a significant micro-reflection was introduced as a mechanism to allow legacy present which was beyond the modem’s DOCSIS 1.0 and 1.1 CMs—which only ability to successfully transmit 16QAM using support the Time Division Multiple Access its 8-tap equalizer, then the operator was (TDMA) protocol—to coexist with the forced to configure the entire channel (and Synchronized Code Division Multiple Access therefore all modems on that channel) to (SCDMA) protocol introduced with DOCSIS utilize the QPSK modulation. Now, if that 2.0. That is, logical channels were created out micro-reflection characteristic was only of necessity to support the large, existing 1.x present for 1% or 2% of the CM population CM base already fielded. However, this on the channel, then we have a scenario in feature supports so much more than just which the operational throughput of the legacy operation. channel is being dramatically limited by a small fraction of the modem population. The Before exploring the benefits of logical logical channel feature allows us to channels, a clear definition is warranted. circumvent this problem by defining up to DOCSIS defines a logical upstream channel four logical channels on that one physical as, “A MAC entity identified by a unique channel. channel ID and for which bandwidth is allocated by an associated MAP message. A For example, logical channel 1 might be physical upstream channel may support configured with a QPSK modulation and the multiple logical upstream channels. The 1-to-2% of poor perform associated UCD and MAP messages ing modems would completely describe the logical channel.” be assigned to that logical channel. Similarly, logical channel 2 could be configured for 16QAM and the remaining population would Logical channels are a unique set of be assigned to this logical channel. Using this transmission characteristics that are dealt with one configuration change, we have now via the use of different modulation (burst) reclaimed nearly a 98% increase in the profiles in the upstream direction. All Logical upstream throughput. Another interesting fact Channels (modulation profiles) are time is that we realized a significant increase in division multiplexed into the same physical throughput by only leveraging DOCSIS 2.0 DOCSIS channel. A physical DOCSIS functionality within the CMTS. That is, we channel is defined by such parameters as: a) realized this improvement even with only 1.0 symbol rate (six rates from 160 ksym/sec to CMs present on the plant. It is easy to see 5.12 Msym/sec in octave steps), and b) center how additional logical channels might be frequency. Logical channels further define the created which leverage higher-order channel with burst profile attributes. modulations (32QAM, 64QAM, and even proprietary 256QAM). For these logical

channels, only the corresponding modems MANAGING IMPAIRMENTS which support such capabilities would be assigned to the logical channel. DOCSIS 2.0 has opened opportunities in which increased efficiencies and greater throughput can be achieved within the return While changes in modulation type path. However, increasing efficiencies and provide for the most dramatic changes in enhancing throughput is not just a simple realized throughput, the use of logical matter of enabling the new features in channels is not limited to this single DOCSIS 2.0. parameter. In cable plants where micro- reflections or amplitude roll-off are significant issues, the use of preamble lengths may be One must first understand the dominant exploited to yield improvements in characteristics and impairments of a given throughput. For example, modems return path before channels can be encountering greater linear distortion could be reconfigured accordingly. When fairly simple assigned to logical channels which utilize characterizations are performed, dramatic longer preambles and therefore yield better increases in throughput can be achieved. equalizer performance, while modems encountering less distortion would be Channels which would be unusable with assigned to logical channels utilizing shorter DOCSIS 1.0/1.1 can now be reclaimed, and preambles. A similar technique can be applied throughput of legacy channels can be to FEC [both codeword length (K) and increased 50% or more with an optimal number bytes corrected (T)], byte configuration. Further, the characterization interleaving, guard times, or any combination methodologies presented in this paper will thereof. This provides the needed mechanism yield relationships between actual plant to deal with the variances of modem signal devices and dominant impairments present quality resulting from such factors as system within a given return path. By identifying loss, amplifier cascades, and micro- dominant impairments and actual plant reflections. devices causing such impairments, this methodology supports targeted maintenance Cable operators can leverage logical activities for so-called low-hanging fruit channels to improve the throughput of legacy improvements that yield major performance DOCSIS 1.0, 1.1 CMs as well as optimize benefits. performance for DOCSIS 2.0 and even 3.0 CMs. They can implement logical channels on The higher bit rates achieved with higher pure DOCSIS 2.0 or 3.0 deployments, but the modulations come at the expense of greater reality is that most cable networks also consist Modulation Error Ratio (MER) requirements of legacy DOCSIS 1.0 and 1.1 CMs. Logical [sometimes incorrectly referred to as Signal- channels can support mixed-mode operation to-Noise Ratio (SNR)] on the upstream and they can optimize the performance of channel. These requirements go beyond the diverse CMs deployed throughout the access traditional first order issues, such as: network. • Thermal noise

• Ingress noise • Impulse noise understand how to automatically compensate This paper focuses on the topic of for linear and non-linear impairments. increased modulation levels and the issues to be overcome to support such levels. Specific With the release of DOCSIS 3.0, issues that will be discussed within this paper spectrum management will become even more are: critical because the CMTS will be faced with maintaining quality of service on multiple • Linear impairments bonded upstreams. Multiple upstream channels will require that MSOs reclaim more • Non-linear impairments and more of their upstream frequency spectrum, including regions which have ADVANCED SPECTRUM MANAGEMENT historically been avoided due to their greater susceptibility to various impairments. Maintaining quality of service across many A spare receiver architected into a CMTS service flows across multiple bonded module can allow cable operators to best upstream channels can not be performed understand the impairments on the DOCSIS manually by an operator and will require infrastructure. It runs in parallel to the live advanced spectrum management to make sure ports to monitor performance of any one of proper flows are assigned to physical channels the upstream ports without materially capable of meeting quality of service impacting the subscriber experience. requirements. With that stated, it is vitally important that the fundamental building The receiver is connected in parallel with blocks for a more detailed analysis that will a selected receiver port so the operator can be required for the future DOCSIS 3.0 be first measure traffic and performance in real time proved out in all DOCSIS 2.0 services for the on any given receiver port. The parallel cable operators today. receiver can access all of the mapping information as well as a full list of cable LINEAR IMPAIRMENTS modems available to whichever receiver port is currently being evaluated. Linear impairments refer to a class of impairments that are signal-dependent and Therefore, while the receiver port being largely unique to a given responding CM monitored is performing its function at full because its transmission path through the capacity, the parallel receiver can non- return path network possesses its own micro- obtrusively gain access to all of the return reflection (impedance mismatch). Moreover, nodes connected to one of the receiver ports the number of amplifiers in cascade also and perform tests on each upstream channel to impact the amount of amplitude distortion and assess its quality and take the time required to group delay (phase) distortion that a CM complete detailed, coherent MER signal will be impacted by, that is to say the measurements. more amplifiers in cascade the more diplex filters the CM signal must traverse. This Cable operators can leverage advanced simply means that the effects of a linear spectrum management to optimize the impairment can only be observed while in the performance of cable modems and better presence of a signal. The signal required for evaluation of linear impairments can either be

very expensive test equipment, or the cable classic RF network analyzer such as the operator can simply opt to use a very Agilent 8753ES (75 Ohm network analyzer) inexpensive DOCSIS CM as the source and a to report the definitive amplitude and group spare receiver architected into a CMTS access delay response from 2 MHz to 52 MHz. module as the measurement tool. Alternatively the VSG/VSA combination is actually more useful in that one can not only receive a detailed report regarding the In the end, when it comes down to amplitude and group delay distortion, but one convenience, speed of measurement, and cost, can also receive a report regarding the there can be little question that the definitive MER versus the carrier frequency combination of a DOCSIS CM and DOCSIS as well. With lab characterization accuracies CMTS is the most effective characterization set aside, the realities of characterizing a live tool available to the cable operator. CATV node in the field poses a variety of Obviously, the measurement accuracy of a issues that lab equipment simply cannot deal system using a DOCSIS CM and DOCSIS with such as: CMTS is considerably less than using a Vector Signal Generator (VSG) such as the Agilent E-4438C or Arbitrary Waveform • Physical distance Generator (AWG) such as the Agilent E5182A and a Vector Signal Analyzer (VSA) • Ingress noise such as the Agilent 89640A, 89650S or N9020A or a second generation CATV Analyzer such as the Sunrise Telecom AT- • Impulse noise 2500RQ, but the speed of the DOCSIS CM/DOCSIS CMTS system more than makes • Live traffic up for the accuracy limitations. Consider the fact that the CM is already installed in the network and that even a detailed DOCSIS 2.0 • Time Synchronization CMTS – CM measurement takes less than 100 ms and a less detailed measurement takes less Therefore, the DOCSIS CM and DOCSIS than 5 ms. The CATV engineer using any one CMTS system offers the cable operator the of the devices mentioned above will have an only characterization technique that is both average measurement time of no less than 10 non-disruptive to customer traffic and seconds for a less detailed measurement and extremely convenient both from availability the measurement time can easily be over three and from a time to measurement perspective. minutes for a detailed analysis. Obviously, one isn’t even discussing the time for a CATV technician to get to a remote site location and It is also important for the cable operator connect the VSG or AWG to the CATV to understand why until recently, linear network. impairments have not been a concern. This is due to the fact that QPSK modulation has no amplitude modulation associated with it, and If a cable operator wants a definitive it is fairly immune to linear distortion affects. linear characterization of a return node This is fundamentally due to the typical cascade of amps and the optical link, then micro-reflection tending to be in the 15 dB to nothing substitutes for setting the entire 25 dB range and given that QPSK needs only circuit in a lab environment and using a an MER > 14 dB to be perfectly acceptable • Amplitude tilt or slope from a performance perspective, one can – coaxial cable easily see why this linear impairment has not been a show stopper for QPSK modulation. Another more important point regarding all NON-LINEAR IMPAIRMENTS linear impairments is that the impact on MER performance is also a function of DOCSIS As in the case of linear impairments, channel bandwidth. system non-linearity is also a signal- dependent distortion. Moreover, while linear That is to say the significance of a micro- impairments such as micro-reflections and reflection or even the effects of multiple diplex filter effects can be observed in the diplex filters has a dramatically larger impact way the noise floor is shaped; there is by on a channel bandwidth of 3.2 MHz than it definition no system non-linearity without the ever did for the older 1.6 MHz bandwidth presence of the signal. In essence, if there is services. One can then understand that linear no signal there is no system non-linearity impairments are a major impact on the wider occurring. More importantly, the impact on a DOCSIS 2.0 channel bandwidth of 6.4 MHz. DOCSIS signal that a system non-linearity This is the primary reason why the DOCSIS presents is also a function of the level of 2.0 Equalizer was increased from 8 taps to 24 QAM that is being transmitted. Simply stated, taps in length. As a result, linear impairments since system non-linearity is signal have only recently generated attention as dependent, then it also holds true that the more and more operators have moved to a 3.2 larger the signal power the more severe the MHz bandwidth first and are now moving to system non-linearity impacts the DOCSIS 16QAM modulation, which with the signal. combination of wider bandwidth and 16- QAM modulation is more susceptible to these Traditionally, most users rely almost affects. As operators seek to enable more entirely on MER to understand channel advanced modulations of 32QAM and quality and predict performance under various 64QAM provided by DOCSIS 2.0 and utilize configurations. Because of the fact that non- the widest channel bandwidth available of 6.4 linearities impact higher levels of QAM more MHz, these impairments will become even so than lower levels, an operator must be more critical. careful when interpreting the MER (SNR) of a communications signal at say QPSK so as not As far as the HFC plant is concerned, to quickly extrapolate the capabilities of that linear impairments generally fall into one of channel to support higher levels of QAM. three classes: Given that system non-linearity has a • Micro-reflections – much greater impact on a higher power signal, impedance mismatches it also follows that system non-linearity impacts the outer points of the transmitted

signal constellation more than the inner points • Amplitude and group of the same transmitted constellation. This is delay distortion – fundamentally due to the fact that the outer diplex filters points are transmitted at a significantly higher power than the inner points and thereby are

impacted significantly more than the inner exhibiting both 2nd and 3rd Order Distortion points. This, too, implies that an operator that occurs on the coaxial cable path that is must be careful when interpreting the common to both forward and return path meaning of MER values reported for a directions. communications channel. CPD is well understood by the CATV For example, if a communications industry in general. It is the phenomena of a channel is configured to transmit 64QAM in a coaxial connector becoming or temporarily nearly perfect communications channel except acting as a diode. It is easily observed by for non-linearity, then all but the outer corner seeing analog video carriers spaced 6 MHz points of the constellation will be nearly apart throughout the return path. While CPD perfect. However, depending upon the degree is easily detectable, the return laser being of non-linearity, the outer points may be so either clipped or just becoming marginally corrupted that they are compressed to a point non-linear can only be witnessed today by of being non-distinguishable from inner advanced spectrum management on a points; that is, a very high error rate may still dedicated receiver on a CMTS card, or by result. For example, if this non-linearity only deploying vector signal analyzer test impacts the 3 outer constellation points at equipment or second-generation CATV each corner, then only 12 points (3 in each of analyzers with a DEMO function and a known the 4 corners) of the total 64 are impacted. reference signal installed on the network. The impact on MER is 12/64 or 19%, which implies that it will have only a minimal With advanced spectrum management, impact on the reported MER. one can easily observe that the effect of any non-linearity is that the outer constellation There is a significant difference in impact points are impacted far greater than the inner to a DOCSIS transmitted constellation when it constellation points. is subjected to a 2nd Order Inter-modulation Distortion (IMD)—commonly referred to in AUTOMATING LOGICAL CHANNEL the CATV Industry as Composite Second CONFIGURATIONS Order (CSO)—versus a 3rd Order IMD Distortion—commonly referred to as Logical channels allow cable operators to Composite Triple Beat (CTB). It is in fact the realize higher upstream bandwidth capacity 3rd Order IMD Distortion that is far more per port by enabling different modulation damaging to the DOCSIS transmitted profiles on a single port. constellation since it impacts the outer points more significantly than the inner constellations, while 2nd Order IMD tends to Proactive configuration tools are required impact all of the constellation points evenly. so that cable operators can implement proactive maintenance strategies that enable automated decision making by providing The system non-linearity that we have actionable data. been referring to is created by two completely different circuit areas of the CATV Network. Common Path Distortion (CPD) is the result Much of the necessary data required to of dissimilar metals acting as a diode and increase bandwidth and perform plant maintenance is captured in the CMTS; MER (SNR)

Logical 40 Logical however, automated configuration Channel 0 Channel 3 management tools are needed so that cable 35 operators can proactively allow automated 30 25

Logical 20 decision-making that configures CMs for dB Channel 2 15 Logical maximum performance based on rules and Channel 1 policies determined by the cable operator. 10 5 0 Cable Modems Now that we have quantified the improvements that can be realized using This management tool could then be used to logical channels, we now turn to the issue of predict the throughput performance of a new automated configuration. That is, how does an configuration, illustrated in the following operator easily identify what logical channel figure. configurations to define and which modems should be assigned to each logical channel. A software tool which supports the collection and sorting of various CM performance statistics is necessary.

DOCSIS 2.0 introduced requirements for a multitude of new performance statistics that are unique to each modem within the plant. Specifically, per modem statistics were added for MER, micro-reflections, and FEC statistics. The number of statistics in combination with the hundreds of thousands of modems makes for a large and complex Logical channels offer huge opportunities data-mining problem that can not be solved to reclaim throughput efficiencies on the manually by a human operator. By extracting upstream. While the magnitude of these modem statistics, a tool could allow the improvement is dependent upon the operator to sort and analyze the distribution of characteristics of the cable plant, these modems on the plant and their associated significant benefits are even available to plant performance measures. The following figure environments which are dominated by 1.0 provides an example of a sample distribution. CMs. The key roadblock to leveraging this This distribution could then be used to isolate technology is the need for automated tools groupings of CMs with similar characteristics, that facilitate the cumbersome task of and logical channels meeting the needs of assigning each individual modem to the each grouping could then be created. appropriate logical channel. However, as the benefits of this capability are realized within the industry, we fully expect vendors to begin offering tools to automate such a process.

Configuration management tools can serve as GUI-based front-ends for CMTS platforms, and they can leverage the insights

gained from advanced spectrum management automate configuration of CMs to maximize to automate the configuration of CMs to performance as well as the option to only maximize performance. allow the system to reconfigure CMs after an operator has manually accepted the Operations personnel could then benefit reconfiguration recommendations. from a logical channel analysis user interface that graphically presents the salient SUMMARY performance information necessary for configuring logical channel operations. DOCSIS 2.0 support for logical channels allows cable to provide significant amounts of Logical Channel Analysis User Interface increased upstream bandwidth. Operators can quickly deliver increased upstream performance to HFC networks consisting of Modem List Window mixed mode DOCSIS 1.x and 2.0 cable Modulation Profile Window modems as well as for new networks Logical Channel Config consisting entirely of DOCSIS 2.0 cable Window

Channel Performance modems. Comparison Window Cable operators can implement logical channels in both mixed-mode and pure Motorola Internal Use

MOTOROLA and the Stylized M Logo are registered in the US Patent & Trademark Office. All other product or service names are the property of their respective owners. © Motorola, Inc. 2005 DOCSIS 2.0 environments and they can They will be able to easily analyze concurrently deploy advanced spectrum logical channels and monitor the automated management to utilize a spare receiver to non- configuration of logical channels. Graphing obtrusively measure spectrum impairments, capabilities allow operators to quickly predict the dominant impairment in any area identify problem CMs and the corresponding of the spectrum by analyzing signal-to-noise percentage of the total modem population measurements, and qualify any area of the impacted. Operations personnel will be able to return path spectrum so they can achieve the review summary update of CM profiles while maximum data signaling rate for logical gaining the flexibility to drill down into channels. detailed windows that present full modulation profiling information. They can leverage logical channels and advanced spectrum management to increase Analysis features provide a graphical the aggregate bandwidth. Automatic logical presentation of the total channel capacity for channel configuration complements and adds both the current profile configuration and the value to advanced spectrum management, and proposed profile configuration. Cable it allows cable operators to successfully operators can allow the configuration manager optimize the highest throughput of a given to automatically initate reconfigurations of channel and better support the QoS demands profiles and move modems to logical channels of real-time services such as voice and video. while retaining the option to manually accept or reject reconfigurations. This powerful management tool therefore provides cable operators with the ability to

DELIVERING ECONOMICAL IP-VIDEO OVER DOCSIS BY BYPASSING THE M- CMTS WITH DIBA Michael Patrick Motorola Connected Home Solutions

Abstract CMTS core, DIBA is an architecture for delivery of high-bandwidth entertainment DOCSIS IPTV Bypass Architecture video over DOCSIS to the home for a price (DIBA) DIBA refers to any of a number of per program that matches the cost for techniques whereby downstream IPTV traffic conventional video over MPEG2 delivery to is directly tunneled from an IPTV source to a set-top-boxes. downstream Edge QAM, “bypassing” a DIBA is an innovative bypass DOCSIS M-CMTS core. It will allow architecture in which the conventional operators to deliver economic IP video traffic Modular-CMTS architecture is replaced by a over DOCSIS infrastructure. hybrid architecture consisting of an integrated CMTS and DOCSIS External Physical INTRODUCTION Interface (DEPI) or MPEG Edge QAMs. This paper/presentation will demonstrate the MSOs are considering IP-video and IP effectiveness of DIBA, discuss how operators Television (IPTV) to supplement their current can evolve their infrastructure to implement digital video delivery. IP-based video enables DIBA, discuss alternative bypass new video sources (the Internet) and new encapsulation protocols to tunnel traffic to video destinations (subscriber IPTV playback either DEPI or MPEG Edge QAMs, and devices). outline the economic advantages of DIBA Of course, such a transition to IPTV migration. It will explain how DIBA removes requires significant additional downstream the boundaries of delivering video using a M- DOCSIS® bandwidth. In a reasonable 5-7 CMTS, and how operators can accelerate year “endgame” scenario, VOD is expected to service velocity by efficiently delivering require 26 times the bandwidth of High Speed IPTV services that bypass M-CMTS Data (HSD). What’s more, attempting such an platforms. increase in downstream bandwidth with a conventional Modular-CMTS (M-CMTS) will THE CHALLENGE OF DELIVERING IPTV be expensive. As an alternative, we propose that IP- IP-video and IP Television (IPTV) create video content travel directly to an Edge QAM, opportunities to supplement their current bypassing the M-CMTS core altogether and digital video delivery. IP-based video enables saving local switch and CMTS core capacity. new video sources (the Internet) and new This technique is called the DOCSIS IPTV video destinations (subscriber IPTV playback Bypass Architecture (DIBA). DIBA refers to devices). It should be considered as a delivery any of a number of techniques whereby option of Video on Demand (VOD). downstream IPTV traffic is directly tunneled The extent to which VOD is delivered as from an IPTV source to a downstream Edge IPTV governs the extent to which additional QAM, “bypassing” a DOCSIS M-CMTS core. downstream DOCSIS bandwidth must be By bypassing the high-cost components of the provided for IPTV. Consider a hypothetical 5- 7 year “endgame” scenario where each video subscriber can download “What I Want When I Want” (WIWWIW). Assume a fiber node of AN INTRODUCTION TO DIBA 750 homes passed has 75% video subscriber penetration, and during the busiest video But there is a simple alternative for IPTV viewing hour 50% of the video subscribers support with a M-CMTS: IPTV should be require an average of 10 Mbps VOD content. tunneled directly to the EQAM, bypassing the The total VOD content required to be M-CMTS core altogether. delivered to that single fiber node during the DIBA refers to any of a number of busy hour is thus 750 * .75 * .50 * 10 Mbps = techniques whereby downstream IPTV traffic 2.8 Gbps, or 73 carriers of 6 MHz QAM 256. is directly tunneled from an IPTV source to a What fraction of this “endgame VOD” downstream EQAM, “bypassing” a DOCSIS bandwidth is IPTV over DOCSIS and what M-CMTS core. By bypassing the high-cost fraction is traditional Digital Video to Digital components of the CMTS core, DIBA is an Set-Top Boxes (DSTBs) is anybody’s guess. architecture for the delivery of high- VOD bandwidth swamps High-Speed bandwidth entertainment video over DOCSIS Data (HSD) bandwidth. Last year, most MSO to the home for a price-per-program that deployments offered only 20 Mbps of HSD matches the cost for conventional video over per 750-household fiber node. A reasonable MPEG2 delivery to set-top-boxes. IPTV is by endgame scenario for HSD would provide definition an IP packet with video content 100 Mbps channel bonded service to 50% of from an IP source to an IP destination address. households passed with a 0.25% concurrency, In traditional IPTV, an IPTV server sends for a total HSD fiber node throughput an IP packet to the destination address of a requirement of 750 * 0.5 * .0025 * 100 Mbps subscriber playback device, which we call = 94 Mbps. Thus, under reasonable generically an “IP Set-Top Box” (IPSTB). hypothetical endgame scenarios, VOD is With the M-CMTS architecture forwarding, expected to require 26 times the bandwidth of the IPTV packet is routed through the M- HSD. CMTS core. In a conventional M-CMTS A key goal of the DOCSIS Modular architecture without DIBA, therefore, the CMTS (M-CMTS) architecture was to define IPTV content is forced to make two transits a common Edge QAM (EQAM) architecture through the Converged Interconnect Network that could support a cost-effective transition (CIN) switch that connects regional networks from Digital Video (DV) VOD delivery to with the core CMTS. IPTV VOD delivery over DOCSIS. A The video/IP traffic travels to the M- straightforward strategy of using M-CMTS to CMTS core, then “hairpins” back through the provide IPTV VOD would call for deploying CIN on a DOCSIS External Physical Interface M-CMTS EQAMs for DV-VOD first, and (DEPI) pseudo-wire to the DEPI EQAM. If- adding M-CMTS core capacity as needed to and-when IPTV becomes a significant transition the DV-VOD to IPTV-VOD. As we fraction of overall bandwidth delivered to a have seen, though, the M-CMTS core fiber node, this hairpin forwarding of IPTV capacity used for IPTV-VOD will quickly content will require significant expenditures exceed the CMTS capacity used for HSD. A by MSOs for M-CMTS core and CIN straightforward implementation of M-CMTS switching bandwidth. for IPTV, therefore, will require many times In DIBA, rather than pass through the M- the cost of M-CMTS core capacity as that CMTS core, the high-bandwidth video/IP currently used for HSD. content is tunneled from the IPTV server,

through the MSO’s converged interconnect bypass traffic and omit these functions. With network directly to the DIBA EQAMs. The an integrated CMTS and no timestamps in the IPTV emerges from the DIBA EQAM with un-synchronized channels, the DOCSIS full DOCSIS framing, suitable for forwarding Timing Interface which is required in the M- through a DOCSIS cable modem to home IP CMTS architecture is not necessary in DIBA. devices. This architecture will deliver high- In DIBA, a DOCSIS 3.0 cable modem bandwidth entertainment video/IP/DOCSIS to requires only one synchronized channel from the home for a price-per-QAM that matches the existing integrated CMTS to provide that of the most advanced, cost reduced timing, control the upstream transmissions, MPEG2 EQAMs. and provide the other MAC functions. The DOCSIS 3.0 cable modem will receive REDUCING DELIVERY COSTS video/IP on additional un-synchronized, inexpensive, DIBA-generated channels, which Operators can deploy independently can even be bonded. scalable numbers of downstream channels without changing the MAC domain or the LEVERAGING EXISTING EQAMS number of upstream DOCSIS channels. These downstream channels are available for DIBA offers a variety of bypass VOD/IP and switched-digital-video/IP. They encapsulations. Using a standard M-CMTS can also lower the cost to deliver video over DEPI EQAM, two bypass encapsulations are DOCSIS service to be competitive with possible, depending on the video server and today’s MPEG VOD. DIBA accomplishes the MSO network. In either case, the server goals of the M-CMTS without the originates a Layer 2 Tunnel Protocol Version unnecessary expense of the M-CMTS. DIBA 3 (L2TPv3) tunnel to the DEPI EQAM. The avoids the expense of the DOCSIS MAC tunnel payload can be: domain technology for the video/IP traffic by using both synchronized and unsynchronized DOCSIS downstream channels. The • The DOCSIS Packet Streaming synchronized channels pass through the Protocol (PSP) in which the video/IP integrated CMTS or the CMTS core and is encapsulated into DOCSIS MAC provide the many DOCSIS MAC functions, frames. This permits the EQAM to including: mix both IPTV traffic originated from the IPTV server with non-IPTV (e.g. Conveying the DOCSIS timestam • ps HSD or VOIP) traffic originated from • Managing ranging to provide the the M-CMTS core on the same proper time-base to the cable modem DOCSIS downstream carrier. • Instructing the cable modems when to transmit upstream • The DOCSIS MPEG Transport (D- MPT) layer which consists of an • Delivering other MAC layer messages MPEG-2 Transport Stream of 188 byte for cable modem registration, packets. All DOCSIS frames, maintenance, etc. including packet-based frames and In contrast, the unsynchronized DOCSIS MAC management-based frames, are channels are generated by the EQAMs included within the one D-MPT flow. (including installed MPEG EQAMs) for the The EQAM searches the D-MPT payload for any DOCSIS SYNC Switched broadcast IPTV multicast is sent messages and performs SYNC without BPI encryption on non-synchronized corrections. It then forwards the D- downstream channels to DOCSIS 3.0 cable MPT packet to the RF interface. The modem IPTV devices. intent of D-MPT mode is to allow MPEG packets to be received by the THE ROLE OF THE CONTROL PLANE EQAM and forwarded directly to the RF interface without having to While there is no design yet for the DIBA terminate and regenerate the MPEG control plane, there are several other control framing. The only manipulation of the planes that must be considered in that design. D-MPT payload is the SYNC The current MSO switched broadcast correction. architecture is intended to save bandwidth for A standard MPEG2-Transport Stream a moderate-sized optical node, and that (MPEG2-TS) EQAM can also be used. Here feature will have to be retained in any DIBA the video server may transmit an IPTV PSP implementation. The PacketCable Multimedia formatted data packet. A PSP/MPT converter, (PCMM) architecture is intended to provide either attached to or embedded within the CIN QoS within a DOCSIS system, and that networking device, then changes the format function will have to be retained. The M- into an MPEG-2-TS that a conventional CMTS architecture is intended to provide a MPEG EQAM can process. An alternative is scalable number of downstream DOCSIS for the VOD server to directly generate channels managed by an edge resource IPTV/MPT formatted packets that the MPEG manager, and this feature is an essential part EQAM can process. In the case of non- of DIBA. IP Multimedia Subsystem (IMS)- synchronized DOCSIS channels, a non- based IP services will include entertainment DOCSIS Program ID (PID) is used because video as well as personalized and on-demand each D-MPT program would require a video. separate MPEG-2 PID. A required extension for sending multiple programs streams of D- VOD AND SWITCHED BROADCAST MPT to the same downstream QAM channel would be for DOCSIS 3.0 cable modems to be Many cable operators have existing VOD programmed to accept D-MPT-formatted and switched digital video architectures. packets with other than the standard DOCSIS These two architectures have quite different PID. purposes, but are similar in one major way to conventional digital broadcast video. In all TUNNELING BROADCAST IPTV cases the video is brought over the MSO network to the EQAMs as MPEG-2 There are several tunneling options that SPTS/UDP/IP. The EQAM then strips off the can be used. One of these is “Interception”, in UDP/IP encapsulation, multiplexes the which SPTS/UDP/IP encapsulation is material into multi-program MEPG-2 TSs that intercepted by a CIN multicast router and are transmitted over QAM carriers. The set- encapsulated into PSP tunnel. Another is PSP top boxes are able to demodulate the MPEG-2 Tunnels directly from the video server to the TSs and decode the video. Since the set-tops DEPI EQAMs. For bonded IP multicast, a cannot receive video over an IP stack, they “DIBA Distributor” component co-located must be instructed which QAM carrier with a CIN IP multicast router distributes or frequency and which Program ID to “stripes” the IP packet to a pre-configured demodulate. In the case of switched digital DOCSIS 3.0 Downstream Bonding Group.

video, the objective is to make more video for IPTV. In one, the client (video viewing) channels available than there is bandwidth for device does not itself support the QoS in a typical RF spectrum. All the available signaling mechanisms and relies on the video content is carried as a collection of IP Application Server as a proxy. The client multicast sessions, in most cases with all sends a service request, and the Application content available within the hub. Only when a Server sends a service request to the set-top within a fiber node requests a Application Manager. Upon receipt of this particular video title is that title carried to that request, the Application Manager determines node. the QoS needs of the requested service and Statistically, some number of set-tops sends a Policy Request to the Policy Server. will be viewing each of the popular video The Policy Server in turn validates the Policy broadcast titles. Other, less popular titles will Request against the MSO-defined policy rules likely be viewed by only one set-top at a time. and, if the decision is affirmative, sends a Still other less popular titles will not be Policy Set message to the CMTS. The CMTS viewed at all. Thus, the limited bandwidth performs admission control on the requested available to that node will accommodate a QoS envelope (verifying that adequate great many more available titles than could be resources are available to satisfy this request), carried simultaneously. installs the policy decision, and establishes the service flow(s) with the requested QoS When the control plane receives a request level. from a set-top for a new title, it searches for an EQAM serving that node which has sufficient available bandwidth. This EQAM is M-CMTS EDGE RESOURCE MANAGER instructed to issue an IGMPv3 join to that IP multicast session. In the case of VOD, the The M-CMTS architecture takes the control plane must locate a server with the PCMM architecture one step further in terms desired title. This server will then send the of granularity. Within the PCMM title as unicast MPEG2-TS/UDP/IP to an architecture, it is the CMTS that makes the appropriate EQAM serving the set-top making admission control decision for new flows and the request. In the case of VOD, there can be assigns then to a suitable DOCSIS channel. In thousands of available titles. the M-CMTS, the core CMTS must in turn request QAM resources from the edge PACKETCABLE MULTIMEDIA resource manager.

PacketCable Multimedia will be an IP VIDEO AND IMS important element of the control plane for DIBA. The objective of managing QoS for the There has been much work within the myriad of services to be provided over the Internet community to use IMS to control limited bandwidth of DOCSIS 1.0 and 2.0 video services over IP to offer roaming and networks led to the development of PCMM. two-way features. This has generally not been PCMM uses such elements as the Application for entertainment video. However, the current Server, Application Manager, Policy Server, aim of IMS has become the universal and CMTS. These are all in the headend and deployment of all IP-based services through under control of the MSO. PCMM is only both fixed and mobile networks, regardless of available for IP-based services. There are location. So it is possible that IMS will several scenarios for the operation of PCMM become a suitable vehicle for the deployment of entertainment video as well, particularly application servers to route particular personalized and on-demand video. messages to. This is an aggressive goal, but not an unreasonable one, given the history of and 3. Interrogating-CSCF is another SIP effort that has gone into IMS. IMS began as proxy, this one serving as the interface an effort by telecom carriers in the Third by which IMS messages from the Generation Partnership Project (3GPP) to outside world access the local converge mobile services with VoIP and IP network. The I-CSCF is able to inspect multimedia by using the Session Initiation the user database to determine, for a Protocol (SIP). SIP is an IP-based peer-to- particular IMS terminal, the correct peer protocol used to establish and control Server-CSCF to forward signaling two-way flows carrying voice, video, and messages. gaming. The actual flows use Real-Time Protocol (RTP) and UDP/TCP. True to its heritage, non-IP devices such as analog The network also has application servers telephones are supported through gateways. that are linked to the control plane via the Server-CSCF. In the past, these applications Gradually the range of services and have been call related, such as call-waiting, access technologies increased, resulting in a call-forwarding, conference calls, voicemail, powerful Next Generation Networking (NGN) etc. However, it is possible to have these architecture. The access technologies include services include such network-oriented tasks any IP/SIP-enabled devices, including phones, as resource management for VOD sessions. PDAs, computers over DSL, DOCSIS, 802.11, etc. The network control plane is built There is an obvious advantage to using around a Call Server Control Function IMS and SIP to provide entertainment video (CSCF) that comprises the following: services in that the control plane would naturally extend to any IP video sources in the Internet on the one hand, and to a multitude of 1. Proxy-CSCF, which is an initial fixed and wireless video playback devices on control plane element to interact with the other hand. The roaming ability and the an IMS device and acts as a proxy in device presence enable handoff between the SIP processing. Beginning with devices, a necessary component of seamless registration, each IMS terminal is mobility. There are many issues to deal with assigned to a Proxy-CSCF. The Proxy- in integrating the VOD and switched digital CSCF authenticates the IMS device video control and data planes with IMS/SIP. and establishes security. It also authorizes the IMS device to use BROADCAST IPTV OVER DIBA network bandwidth, and may apply QoS policies to this use. It is in the path of all signaling. In the typical MSO approach to switched broadcast video, the set-tops are non-IP so it 2. Server-CSCF, which is a SIP server is the MPEG EQAM that joins the IP and binds the user IP address to its SIP multicast group. The multicast group is then address. It interfaces to the user mapped to a particular QAM and PID, and it database (the Home Subscriber Server is this information that is forwarded to the set- or HSS) where user information is top to enable it to receive the MEPG-2 video kept and sees all signaling messages. It transport stream. also determines which of the

In fact, DOCSIS is not generally used to deliver broadcast entertainment video, since • Data Plane operation from IPTV server the bandwidth is too expensive. However, the to DIBA PSP/MPT Converter, DIBA intention of DIBA is to bring the cost of Distributor, and DEPI/MPEQ EQAMs DOCSIS bandwidth low enough to make

IPTV viable. With DIBA, the IP set-top will be able to join the IP multicast group. The • NGOD Session Manager control switched broadcast control plane determines signalling to IPTV server which programs are multicast to which fiber node, consistent with the IP set-tops receiving • CMTS control signaling to DIBA which programs. Distributor and PSP Converter functions in CIN BROADCAST IPTV TO DOCSIS 2.0 AND 3.0 MODEMS • CMTS control signaling to EQAM Broadcast IPTV is possible with both DOCSIS 2.0 and 3.0 modems. In either case, DIBA will remove the boundaries of the IP set-top box sends an IGMP join to an delivering video using a M-CMTS, and it will IP multicast session for a particular program. allow cable operators to leverage service The DOCSIS 2.0 modem will have to use a velocity by efficiently delivering IPTV synchronized DOCSIS downstream channel, services that bypass the M-CMTS core. while a DOCSIS 3.0 modem can receive on a Bonded Channel Set. It may or may not be ABOUT THE AUTHOR necessary to change DOCSIS channels to change to a new IPTV ‘channel’. However, Mike Patrick is a Data Networking any DOCSIS channel changing requires Architect, Motorola Connected Home communication from the CMTS to the cable Solutions. In this role, He is responsible for modem in the form of a Dynamic Channel the architecture of Motorola’s Cable Modem Change (DOCSIS 2.0) or Dynamic Bonding Termination System products (CMTS). He is Change (DOCSIS 3.0) to change a tuner of currently leading technical development of the the cable modem to the new channel. DOCSIS® 3.0 feature set for Motorola’s BSR 64000 CMTS/Edge Router. SUMMARY Mike joined Motorola in 1994. Over the last 13 years, Patrick was a Data Networking Cable operators will be able to System Engineer and was a major contributor economically supply IPTV capacity of to the Downstream Channel Bonding and IP gigabits per fiber node by bypassing the M- Multicast features of DOCSIS 3.0. Mike was CMTS core and tunneling IPTV directly to a major contributor to version 1.1 of the Data the M-CMTS EQAM. DIBA allows operators Over Cable System Interface Specification to avoid the cost of additional M-CMTS core (DOCSIS), which standardized QoS operation capacity for IPTV. DIBA is proposed as a for VoIP cable modems. work item for CableLabs after DOCSIS 3.0 Before joining Motorola, Mike served in draft specifications. If it is approved, various engineering and management CableLabs would standardize: positions at Digital Equipment Corporation, General Computer Corporation, Cryptall Corporation, Ztel, and Texas Instruments. He has 14 U.S. patents granted to date. Mike earned his bachelor’s of science and master’s of science, computer science degree from the Massachusetts Institute of Technology in 1980, as well as a Masters in the Management of Technology from MIT Sloan School in 1992. He can be reached at [email protected]. Enhancing Switched Digital Video for Support of Next-Generation Advertising through Switched Digital Programming By Joe Matarese, SVP, Advanced Global Technology C-COR, Inc.

Abstract In order for SDV to take advantage of specific program knowledge and for SDP to Several years into the development, and a be deployed, several things must happen: little over a year into the deployment of Switched Digital Video (SDV) technology, is Modeling of SDV narrowcast bandwidth it possible that the cable industry continues requirements must be redone to make use of to overlook much of SDV’s promise? This program-specific audience measurement paper considers a specific aspect of SDV data, perhaps from Nielsen Media Research. where cable providers have the ability to make significant advances in delivering SDV systems must be instrumented to value to programmers and advertisers, as generate audience measurement data from well as the opportunity to monetize such real time collection of channel change value. We define this aspect as Switched signaling. Digital Programming (SDP). SDV systems must be enhanced to facilitate SDP is characterized by a focus on explicit program joins and leaves, either by a optimal design of switched digital video human viewer or by a DVR, so that program- systems to deliver programming, not simply specific audience measurement does not channels, in the most bandwidth efficient incorporate artificial “channel loitering”. manner. Existing models of SDV typically consider channel popularity in determining This paper will investigate each of these the amount of narrowcast bandwidth that undertakings in detail and provide statistics one must dedicate to delivery of the SDV from an actual field-trial of SDV in order to channels. While reliance on channel build these measurement models. popularity information is useful in obtaining a rough estimate of SDV sizing, to get a more As a consequence of deploying SDP, accurate reading one must consider the programmers will benefit in addition to actual programs. For example, while MSOs. Through the MSOs programmers can Network A might have a general popularity obtain accurate, “raw” audience rating, averaged over the week, of six points measurement data on which to base program audience share, it is clear that some lineup decisions. Moreover, programmers programs offered by the network will have have the potential to deliver multiple network much higher audience shares while other variants, possibly targeted at different programs will have lower audience shares. demographic profiles, rather than deliver a A similar phenomenon is seen with VOD single “one size fits all” network that tries to programs. The relative popularity of a satisfy all viewers. specific program, averaged over its entire license window, is often significantly less than the relative popularity of the same program averaged over a shorter period like the prime time viewing window or the new release window. INTRODUCTION of SDV in both modes requires that the goals of next-generation advertising be made a With new advertising models much on the vital component of SDV applications minds of cable industry strategists, it’s planning. important that strategic planning respecting use of switched digital video technology take Several considerations enter into an into account the great benefits to be derived assessment of what must be done to optimize from SDV for advertising purposes. the SDV environment for enhanced advertising, including: Over the past year the case for development of new advertising models has Adjustments in SDV architecture that allow moved to the industry’s front burner amid operators to move away from switching signs of dissatisfaction from Madison based on channel demand as measured over Avenue over ROI performance of television long time segments to a per-program advertising in comparison to advertising on switching capability based on close tracking the Internet. Operators recognize cable’s of recent individual program performance VOD and SDV infrastructure opens new over short time intervals; opportunities to extend the power of the TV medium to advertisers at levels of Implementation of data-collecting addressability and viewership assurance capabilities to enable this SDP paradigm, never seen before in TV advertising. which will require generation of audience measurement data from real-time collection But to fully exploit the advertising of channel-change signaling; benefits of this interactive infrastructure operators may need to go farther than they Agreement on a process that facilitates have to date to ensure the capabilities are in system recognition of explicit program joins place that will allow addressable advertising and leaves by both viewers and DVRs so that to become a major revenue center for the program-specific audience measurement future. This may add marginal costs to SDV does not record “channel loitering” as actual infrastructure costs, but these will be more viewing; than offset by the contribution a robust new advertising revenue stream makes to the Modeling of advertisement placement and return on the overall investment in that SDV the underlying support architecture around infrastructure. demographic profiles as opposed to geographic zoning. Presently, planning surrounding implementation of addressable advertising Switched Digital Programming models in video-on-demand and other digital service applications is pursued as a largely SDP is a method of using switched digital separate endeavor from the planning video systems to deliver programming, not associated with implementation of switched simply channels, in a manner that’s digital broadcast and unicast capabilities. optimized for addressable advertising. The latter are primarily tied either to Implementing switching at this level of bandwidth savings, in the case of switched granularity gives operators an opportunity to broadcast, or to time-shifted delivery of maximize the placement of ads based on the broadcast programming in the case of unicast profile of each individual viewer of programs models. But full realization of the revenue- with the lowest viewing levels while making generating and operations-savings potential sure the programming with the most appeal to multiple viewers is always in the Each chart reflects how an 85-channel broadcast tier. lineup might be delivered digitally to a 500 tuner user group through a combination of Existing models of SDV typically broadcast and SDV multicast. In this model determine which programming belongs in ten channels are allocated to broadcast, the SDV channel cluster on the basis of leaving the remaining 75 networks to be measurements of network popularity over delivered via SDV based on viewer relatively long periods of times. But while selections. Studies show that at peak hours reliance on channel popularity information is the percentage of viewers tuned to SDV useful in obtaining a rough estimate of SDV averages about 14 percent of the viewing sizing, to get a more accurate reading one population. In the charts shown here the must consider the actual programs. focus is on the number of unique channels being watched via SDV through the course For example, while Network A might of the day (shown in purple with the highest have a general popularity rating, averaged peak) and the number of unique tuners that over the week, of six points audience share, are engaged with SDV-delivered it is clear that some programs offered by the programming. network will have much higher audience shares while other programs will have lower Ideally, in terms of the benefits of using audience shares. A similar phenomenon is individually targeted advertising, the seen with VOD programs. The relative operator will want the number of SDV popularity of a specific program, averaged programs being watched to sync precisely over its entire license window, is often with the number of tuners that are viewing significantly less than the relative popularity those programs. When that happens there is of the same program averaged over a shorter just one viewer per SDV-delivered program. period like the primetime viewing window or the new-release window.

A compilation of SDV data from field experiences based on three different criteria for setting switching priorities illustrates the benefits of SDP. These measurements, as shown in Figures 1, 2 and 3, were taken across a service group consisting of 500 tuners. SDV Utilization forSDV 85 channels Utilization across 500 tuners [Broadcasting 10 most popular channels] 80 70 60 50 40 30 20 10 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 Hour of Day

Unique Tuners Unique Channels

Figure 1 SDV utilization for 85 channels across 500 tuners (broadcasting 10 most popular channels)

Figure 1 assumes the choice of hours but also on either side of the peak programming to be broadcast over the ten times. In other words, more than one tuner is broadcast channels is based on network tuned to a significant percentage of the SDV popularity as reflected in viewership over the programming at these times. In Figure 1 the course of the entire day. Here, as shown by blue peak in the evening is tallest of the three the blue area of the chart, there are scenarios, which reflects the fact that there substantially more tuners watching SDV are some popular programs airing in peak programming than there are programs being hours that are on otherwise less popular delivered in SDV mode, especially at peak networks. SDV Utilization forSDV 85 channels Utilization across 500 tuners [Broadcasting 10 most popular channels at busy hour] 80 70 60 50

40 30 20 10 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 Hour of Day

Unique Tuners Unique Channels

Figure 2 SDV utilization for 85 channels across 500 tuners (broadcasting 10 most popular channels at busy hour)

Figure 2 shows SDV utilization when the turners across the entire day. This pattern ten broadcast channels are chosen based on results because there are some networks that network popularity measured at the peak are very popular over the course of the day busy hour. This lowers the number of but don’t necessarily have the most popular multiple tuners per SDV channel at peak but programs on air during peak hours. spreads the distribution of multiple SDV SDV Utilization forSDV 85 channels Utilization across 500 tuners [Broadcasting 10 most popular programs] 80 70 60 50 40 30 20 10 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5 Hour of Day

Unique Tuners Unique Channels

Figure 3 SDV utilization for 85 channels across 500 tuners (broadcasting 10 most popular programs)

Figure 3 shows SDV utilization when the which means those unique programs are ten most popular programs are broadcast available for ad insertions based on based on program popularity measured at demographic profiles that provide a more each hour of the day. This results in the most granular focus for advertising than is the case effective use of SDV for advertising with geographic- or zone-based ad insertion. purposes. Niche programming is thus optimized for setting higher CPM (cost per thousand) If the SDV system is agile, always avails on either side of the two moving the most popular programs at any or three busy hours than would be the case in given time to the broadcast QAM and the other scenarios. switching the rest, the blue area is mitigated at peak and throughout the day. This The concept of demographic profile- increases the percentage of time that just one based advertising can be an important driver viewer is watching each SDV-delivered behind operators’ planning for use of SDV, program. Thus, while SDV multicast is being especially if SDP capabilities are factored in used in this example, the operator is to maximize the addressable advertising maximizing the amount of programming that opportunities. One illustration of the benefits is effectively being delivered as if it were of using SDV to support an advertising unicast. model based on demographic profiling can be found in a large metro interconnect SDP and Advertising scenario where, today, the niche programming lineup is determined on a As demonstrated by Figure 3, SDP zoned geographic basis. ensures that there's a big chunk of the day where each tuner is on a unique program, Consider, for example, a hypothetical needs for accurate information about ad metro interconnect where there may be 80 viewership. To ensure SDP insertion is zones, each of which is served by 100 niche responsive to the fast-paced changes in channels matched to the zone profile. In this audience viewing preferences, the system scenario, to support addressable advertising requires a very agile feedback loop that just on a zone basis in effect requires the regularly delivers audience information interconnect to be equipped to insert ads across all channels. individually on what amounts to the equivalent of 8,000 channels. Currently available data collection and management systems provide operators the The use of SDV technology in means to gather and categorize such conjunction with demographically information into useful components for determined ad insertion greatly reduces the operations, ad insertion and advertiser insertion scale requirements while creating a performance analysis applications. But it’s much better means of matching ads to essential that the monitoring be applied viewers, thereby raising the CPM value of across broadcast as well as narrowcast such advertising. For example, assume there QAMs to create the full picture of program are eight defined demographic profiles for performance and ad viewing that a SDP advertising purposes and that, on average, system requires. each of the 80 niche channels (or programs in the case of SDP) on view at any given Moreover, from an operational time might need to be transmitted in four perspective, generating reports that look at streams to accommodate the demographic SDV and broadcast holistically is essential to profiles of viewers tuned to that capacity planning, knowing what switch programming across the entire interconnect requirements are, when nodes should be region. (Some programming might have just split, etc. With an agile, comprehensive one demographic viewing profile, while feedback platform, operators will be able to others might have more than four, but this look at programming in 30-minute intervals, can be assumed to be the average.) take the measure of shifts in popularity and determine which programs to move in and In this case, the number of channels the out of the broadcast tier. insertion system must be able to serve goes down drastically. With four channels The compilation of data for advertisers’ required per niche channel, the interconnect purposes as well as for accurate needs to be able to insert ads into just 400 measurement of true viewing patterns on a channels to cover the entire region. Thus, a per-program basis will require that DVR far more advantageous degree of tuners be monitored with an understanding of addressability is achieved in a far more which recorded programs are actually manageable insertion environment. watched and whether ads are skipped. Existing data management systems allow Improving Accuracy of Viewing Data operators to collect such information and verify the amount of time spent watching Introducing the idea of SDP has important recorded programming and the ads. implications for how program viewing is monitored and for the collection of data that Supporting SDP also requires a valid can be extremely useful not only to measure of when a TV tuned to managing bandwidth and determining viewer programming is actually being watched. profiles, but also to fulfilling advertisers’ Such information would also contribute greatly to the advertiser’s understanding of judging ad performance on a per-program program and ad performance. But obtaining basis. this information would require implementation of a mechanism that would All of these contributions to an enhanced require consumers to proactively signal when data-collecting capability for purposes of they are joining or leaving a program. supporting SDP will provide operators the ability to deliver the improved advertising For example, the user interface could be performance analysis advertisers are looking programmed to put a message on screen after for. Exploiting these capabilities will require each program asking the viewer to indicate coordination of efforts between technology whether the next selected program is going and advertising departments and a high-level to be watched. If the answer is no or there is strategic commitment to the potential of no response, the system could switch the next-generation advertising in cable. screen to a barker channel until the viewer signals a program request. Cable operators will need to proactively pursue the support of audience measurement Does the need for accurate viewing entities such as Nielsen to ensure the records information merit implementing a “lean- are compiled in a consistent manner forward” requirement that is intrusive to endorsed by the advertising industry. traditional viewing behavior? One can argue Programmers will have to be persuaded that that the time has come to recognize the the pain of delivering the “truth” about ad potential of advertising as a major performance, which is to say, valid contributor to revenue growth going forward. performance data that accounts for ad As competition intensifies downward skipping and channel loitering, is well worth pressure on subscription prices and as the benefits to be gained from providing addressable advertising provides cable advertisers the ROI they are looking for. operators a way to offer much greater value Fortunately, this may not be too hard a sell at to advertisers, there is reason to expect the a moment when the risks to not delivering balance between ad and subscription more accurate performance data are already revenues will shift toward a greater share on apparent in the growing share of ad dollars the advertising side. going to the Internet.

For example, operators are already The Advertising-Optimized SDP considering the possibility that by giving Architecture viewers the option to view advertising as an alternative to paying full price for premium All the functionalities of an advertising- on-demand programming they could expand optimized SDP infrastructure are available the audience for such programming and for implementation by cable operators, thereby derive higher revenues through the should they decide to make advertising CPM rates charged for the ads. The opt-in revenue acceleration a core driver to SDV potential of advertising-support strategies and data management planning. could be applied to the SDP model as well, where consumers who express a willingness The essential requirements are: traditional to signal their viewing intentions would be silos be eliminated through coordinated given a discount on their subscription price. management of broadcast and narrowcast The percentage of viewers willing to do this QAMs; VOD and SDV resources must be would likely be large enough to comprise a managed from a single platform, and valid statistical base for advertisers to use in advertising and applications policy management must be applied across all - mobile, IPTV, PC or television - from a applications. single server. With the switching architecture in place to support SDP, this device-specific, A unified session resource management user-specific capability puts operators on system ties all this together. It is responsive course to meet market requirements as they to command flows from both the VOD client unfold for years to come. on the set-top and the program selection process within the electronic programming Summary guide. When the subscriber requests a program, whether from the VOD or the SDV The importance of SDV to advertising side, the system identifies the specific server can’t be overstated at this moment of turmoil or switch interface, instigates the subscriber in the TV advertising domain. Advertisers join controls for SDV multicast, registers the are looking for ways to overcome the low bandwidth requirements for a VOD or SDV returns on TV advertising as audiences session and identifies the path through the fragment around niche programming, avoid access network for delivery to the set-top advertising through use of DVR technology box. and spend ever more time away from the TV and on the Web. Advertisers not only seek In instances where core and edge QAMs greater efficiency in reaching desired serve to modulate bit streams onto RF audience segments; they want better frequencies that have been allocated to a information concerning the performance of specific server group, the unified bandwidth their management system selects the appropriate ads. QAM module for content delivery. And, by tracking bandwidth utilization across the Cable operators can greatly expand the QAM and server arrays, it performs the load value of SDV for advertising purposes by balancing essential to preventing bandwidth developing a Switched Digital Programming bottlenecks in the distribution system. capability that ties in with a full accounting of viewer behavior across broadcast, SDV, Moreover the unified platform provides VOD and DVR segments. Demographic the resource management control profiling can be used rather than simple zone mechanisms that are essential to executing segmentation to allow advertisers to reach interactive applications requiring the people they want to reach within each transmission of specialized data content to a program. particular set-top. For example, in an advertising application where a viewer is Highly accurate accounting of individual requesting transmission of a long-form programming and ad viewership enabled by version of an ad, the SRM sets the bandwidth a comprehensive data-collection and requirements and duration for the session by management system allows operators to coordinating bandwidth allocations through provide advertisers an unprecedented level of the policy server and appropriate edge QAM assurance as to the return on their ad dollars. to allow the expanded ad to be transmitted to The network and back-office components are the viewer. available to support the new advertising paradigm. Operators who make advertising By eliminating the platform silos between requirements a key consideration in their use advertising, SDV and on-demand content, of SDV and on-demand infrastructure will be the unified platform also allows operators to well positioned to make advertising an ever stream content and advertising to any device larger contributor to their revenue stream. EXPLOITING MICROCELLS ON CABLE PLANT TO LEVERAGE SPECTRUM ASSETS Stuart Lipoff IP Action Partners

Abstract INTRODUCTION

This paper will describe how cable Following the FCC's AWS auction the operators can exploit their outside plant SpectrumCo consortium of MSOs now owns assets to host wireless microcells to do more approximately 20MHz of spectrum with than compensate for their smaller spectrum nearly ubiquitous national coverage. The 137 assets relative to mainstream cellcos who licenses acquired cover 260.5 million pops at employ macrocell architectures. By means of a cost of $2.4 billion (equates to $0.46/MHz- computations and modeling, it is pop). demonstrated that exploiting microcells can This spectrum is situated between the not only provide competitive parity to existing 800MHz and 2GHz incumbent mainstream cellcos with more spectrum, but cellular bands and is prime "real estate" to in fact a hybrid macrocell/microcell launch wireless services that could be architecture with only 20MHz of licensed employed to enable a "quad-play" service spectrum can actually enhance the ability of offering that would be competitive with the wireless operator to offer leaf-frog incumbent cellular operators (cellcos). services that incumbent cellcos will not be A competitive wireless service for able to duplicate without adding more much cable is not just an upside opportunity, it is a more spectrum. survival imperative as Table 1 demonstrates.

Table 1 Strategic Imperative for Cable Issue Implication

Verizon and AT&T already lead in wireline To defend against churn cable must be able and wireless market share, have significant to offer at least a competitive quad play high speed internet adoption, and are moving bundle rapidly to add IPTV to their mix

The mainstream cellcos have covered their For cable to build a me-too greenfields historical fixed CapEx investment in wireless wireless network with competitive coverage plant and are largely investing success and quality of service will require massive capital fixed capital investment with long payback periods Wireless voice adds already exceed wireline Although cable is realizing impressive growth adds and over time, significant migration from and revenue from wireline services, unless wireline to wireless can be expected the industry has the ability to offer wireless voice it will lose out in the future migration

Cellular is so competitive that voice margins Cable has control of, and experience in are depressed so the mainstream carriers are formulating, highly valued multimedia assets attempting to develop high value added non- that could allow cable to leap-frog over simple voice services to boast ARPU me-too voice service offerings Figure 1 Histograms of Spectrum Holdings of Mainstream Cellcos

Sprint PCS + Nextel Spectrum Holdings in 86 Major Metro Areas tMobile Spectrum Holdings in 86 Major Metro Areas min =24MH z mean=49MHz max =84MHz min =0MHz mean=27MHz max =60MHz 60 60

50 50 New Y ork New Y ork 40 40 City=54MHz City=30MHz 30 30

20 20

10 10 Number of Areas Metro Number of Areas Metro Number 0 0 0-10 10-20 20-30 30-40 40-50 50-60 60-7 0 70-80 80-90 0-10 10-20 20-30 30-40 40-50 50-6 0 60-7 0 70-80 80-90 PCS MHz/Metro Area PCS MHz/Metro Area

Cingular+AT&T Spectrum Holdings in 86 Major Metro Areas Verizon Spectrum Holdings in 86 Major Metro Areas max =65MHz min =10MH z mean=50MHz max =90MHz min =0MHz mean=32MHz 60

60 50 50 New Y ork 40 New Y ork City=45MHz 40 City=65MHz 30 30 20 20 10 10 Number of Areas Metro Number of Areas Metro Number 0 0 0-10 10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 0-10 10-20 20-30 30-40 40-50 50-6 0 60-7 0 70-80 80-90 PCS MHz/Metro Area PCS MHz/Metro Area

Source: computed from RCR News database prior to AWS auctions

Spectrum Required required in the top 20 USA cities (as ranked in order of highest population density). The mainstream cellcos typically hold on the order of 40MHz of nationwide spectrum nearly twice the spectrum won by Figure 3 Spectrum Required to Support 2006 cable in the AWS auctions. As shown in Levels of Cellular Traffic Figure 1. 80 70 Legend Furthermore, as Figure 2 shows, to Spectrum req for voice 60 support the very high population density of Spectrum req for data 50 New York City (NYC) and their >20% share * Based on traffic capacity analysis model developed during this project of market, Verizon is holding 65MHz of 40 spectrum in NYC. 30

Required Spectrum (MHz)* 20 Figure 2 Example Competitive Holdings in NYC 70 10 60 0 NYC Miami Detroit Boston Tampa Buffalo Dayton Norfolk El Paso El Hart ford

50 Chicago Honolulu Salt Lake Cleveland San Diego Milwaukee Providence Washington Philadelphia

40 San Francisco

30

20 SpectrumCo Figure 3 is based on a model to be further 10 described in this paper under the following set of key assumptions: 0 Sprint PCS+Nextel tMobile Verizon Cingular+AT&T • CDMA2000 all IP EVDO-Rev A air interface

In fact as Figure 3 shows, based on • A 25% market share typical of today's rapidly growing voice+data traffic Cingular and Verizon market share load, more than 20MHz of spectrum is leaders • Traffic load of Market Reality

– 142 minutes of use per month of Formulating a plan that optimizes the voice traffic engineered for 2% cable industry assets requires an grade of service blocking understanding of the marketplace realities. In particular what must be understood are the – Additional data traffic based on underlying economics and the implications for 10% adoption of data by voice the current and future traffic load. subs who consume on the average of 25MBytes/month of As Figure 4 shows, the average downstream data revenue per user (ARPU) is flat to slight declining while minutes of use continue to • Macrocell architecture with the grow as a rapid rate. Data traffic today is minimum radius of macrocell of 0.5 minimal since only it accounts for only 10% mile of ARPU and further because most of the data revenue is from very low bandwidth Figure 3 is actually a best case for incumbent messaging services. carriers since they need to waste some spectrum to support legacy, less spectral efficient, air interfaces and manage dual use Figure 4 Historical Trends in Cellular Voice while they transition to new 3G and 4G Traffic Load and Revenue technologies. Cents/ MOU/ ARPU/ Minute Month Month* 250 160 $50.00 Because MSOs could build a 140 $49.00 200 208 ARPU/Month 121200 $48.00 greenfields wireless network that fully Cents/Minute MOU/Month $47.00 exploits 3G+ air interfaces that are more 151500 101000 80 120 $46.00 spectrally efficient than legacy cellco 101000 92 60 $45.00 networks, it would be possible to build out a 73 40 $44.00 50 40 fully competitive voice-centric cellular 54 20 $43.00$43.00 network with comparable voice capacity to 35 0 0 $42.00 cellcos even with one half the spectrum. 2000 2001 2002 2003 2004 2005 2006 * APRU= 90%Voice+ However, it would be desirable to do more 10%Data in than construct a "me-too" wireless voice 2006 service offering. In the short term, a "me-too" service offering would make the acquisition The net result is that the ratio of revenue to of subscribers a slow process that is a highly traffic is rapidly declining. This trend toward undesirable position to be in as a late entry a commodity status for voice demonstrates the competitor in the marketplace. In the longer need to find new non-voice services that will term, cellco competitors would be likely to generate high margin ARPU. use their additional spectrum to launch non- voice services that would make the spectrum starved MSO's cellular network non- Table 2 Revenue and Bandwidth of Over-the- competitive. Air Unicast Non-Voice Wireless Services

Average Capacity Service Bandwidth Revenue Usage Mb/s For both short and long term reasons, SMS Message $0.07 0.0002 it would be highly desirable for the cable MMS Photo $0.20 0.0098 AWS spectrum owners to leap-frog beyond a 2 minute voice call 12 Kbps $0.07 1.4400 me-too voice centric wireless offering and 2 minute low-resolution video 128 Kbps $0.99 15.3600 instead offer a multimedia rich voice+non- 6 minute low-resolution video 128 Kbps $0.99 46.0800 6 minute high-resolution video 384 Kbps $0.99 138.2400 voice service from the start. However it 30 minute high-resolution video 384 Kbps $1.99 691.2000 would also be highly desirable to build this Source: “The Economics of Mobile Broadcast TV”, Yoram Solomon, Mobile DTV Alliance leap-frog network without having to engage in the expensive and uncertain process of acquiring additional spectrum much beyond the 20MHz already in hand. The challenge however is that spectrum well in excess of that held by although subscribers will pay substantial fees competitors. The ideal situation for cable is for highly valued multimedia services, these to execute a strategy that not only builds on highly valued services (e.g. video) generate content assets but also employs a non- very large traffic loads. Examples of non- conventional architecture that leverages the voice traffic are shown in Table 2. installed base of cable plant and technology.

These traffic loads are so intense that Compared to cellcos, the cable that their $/megaByte can not be industry has unique access to a broadband low economically supported on today's cost wired backbone network that can be macrocellular 3G cellular networks at leveraged to provide economically attractive economic levels of CapEx investment in high bandwidth multimedia services. infrastructure as shown in Figure 5. Furthermore, we will show that even with 40- 60MHz of spectrum, incumbent cellcos can not economically scale their macrocellular Figure 5 Revenue Per Megabit from Example networks to support the capacity needed to Non-Voice Wireless Services satisfy consumer's appetite for high bandwidth

Revenue per Megabit from the Different Services multimedia services. On the other hand, the SMS Message $375 MMS Photo $20 cable operators can exploit their extensive 2 min voice call $0.05 2 min 128Kbps video $0.06 HFC network to economically support a * 6 min 128Kbps video $0.02 microcell underlay network that can more 6 min 384Kbps video $0.007 30 min 384Kbps video $0.003 than compensate for the cable operators “Pain Threshold” $0.02 limited 20MHz of spectrum. Source: “The Economics of Mobile Broadcast TV”, Yoram Solomon, Mobile DTV Alliance Not only will we demonstrate the A NATURAL STRATEGY FOR CABLE ability for such a microcellular based underlay network to offer a non-voice traffic capacity Assuming greenfield wireless build- advantage, we will also note that a hybrid out for cable, a leap-frog wireless strategy is called for given cable’s late start and the need macrocellular network combined with a to accelerate subscriber take-up to cover the microcellular underlay will allow additional large fixed CapEx investment needed to build degrees of freedom in the overall architecture a network with competitive coverage. and the associated service offerings that Greenfields spectrum can be built by cable further leverage cable industry multimedia using most advanced spectral efficient interests and assets. technology to gain on the order of a 2X capacity advantage over incumbents who carry legacy technology baggage; however, Multimedia Non-Voice Traffic Loads incumbents will soon catch up as they rebuild to next generation air interfaces. Today’s cellcos offer data plan pricing that top out at about 50MB/month, but even Cable’s natural strategy is to employ modest multimedia traffic in excess of multimedia highly valued video and other entertainment assets integrated with wireline webpage surfing breaks this 50MB budget and high speed data products to create highly and places a non-economic load on their differentiated services and counter declining >40MHz of macrocellular infrastructure. ARPU/minute trends. Although next Some examples of multimedia traffic are generation air interfaces can provide some shown in Table 3. Note that all the examples temporary capacity advantage, it will not be possible to execute a multimedia bandwidth * While microcells on aerial plant are straightforward to add, in hungry leap-frog strategy employing congested business districts with underground plant the costs will be conventional cellular architectures without higher. beyond webpage downloads, break the Study Methodology 50MB/month threshold. Our analysis employed the following steps: Table 3 Illustrative Multimedia Data Traffic Load • Employ the central business districts Media Event MegaBytes Events/Day 20% Busy MB/Month (MB)/Event Hour kb/s (22 days) in New York City as a worst case 6 min 384 kb/s video QVGA 17.28 1 7.68 380 USA example for spectrum needed in (320x240 pixels) at 15 f/s MP3 Song Download 5 5 11.11 550 high population congestion limited eMailFile Attachment Downloads 0.3 20 2.67 132 areas.

We bPage Downlo ads 0 .0 5 5 0 1 .1 1 5 5 • Compute the amount of spectrum required to serve today's traffic load Since building a bottom up forecast for a conventional macro-cellular for consumer's appetite for non-voice data architecture consumption is so dependent on forecasting the adoption of specific services, it is highly • Propose an alternative microcellular speculative to rely upon a bottom up buildup architectures for cable of traffic. A top down approach looking at consumer's appetite for fixed wired data • Re-compute the capacity provided for consumption on high speed cablemodem or each microcellular alternative under DSL residential services can be viewed as a consideration and compare that to high end upper bound for wireless data usage. – Today's voice+data traffic loads Table 4 is based upon today's average – Future possible multimedia Table 4 Average Data Consumption by Fixed intensive data traffic loads Wireline Cablemodem Residential Subscribers

25 Average Wired BB Downstream Busy Hour Data Rate/Sub (kb/s) MACROCELLULAR ANALYSIS 11 Busy Hour Downstream Data Consumption/Sub (Mbytes) 20% Busy Hour Concentration Factor (percent of all daily usage) 1,209 In order to have a baseline to compare Monthly Downstream Data Consumption/Sub/Month (Mbytes) 20% with a recommended microcellular approach Ratio of Upstream to Downstream Data Rate 242 for cable, we begin by analyzing the capacity Monthly Upstream Data Consumption/Sub/Month (Mbytes) of a conventional macrocellular architecture data consumption by cablemodem in the example NYC CBD. subscribers. It shows that the current fixed wired downstream consumption of 1,209 Manhattan Central Business Districts Mbytes/sub/month is 240 to 24 times the 5-50

Mbytes/sub/month consumed by today's The amount of spectrum required in wireless data subscribers. Furthermore if one any one market area is determined by the considers that today's wireless data ARPU is traffic capacity of the most density loaded only 10% of today voice+data ARPU, if data cellsite in that market area. Using the NYC were adopted by a larger percentage of voice market area as an example, the most density subscribers well in excess of what 10% ARPU loaded cellsite can be expected to be in either implies; the total data traffic load on the Lower or Midtown Manhattan areas shown in wireless system could well grow on the order Figure 6 Lower and Midtown Manhattan of 10 times additional. Central Business Districts.

Table 5 Macro Cell Analysis Model for CDB in Manhattan Manhattan Dem ogr aphics Ce llSite Loading 1 Manhattan land area (miles^2) 28.4 input 19 CellRadius (miles) 0.50 input 2 Nighttime population 1,487,536 input 20 CellArea (miles^2) 0.79 calc 3 Daytime Population 3,389,300 input 21 Daytime Pops in low er Manhattan cellsite 572,602 calc 4 Ratio Day/Night 2.3 calc 22 Cellular Adoption 72% Input 5 Mid-tow n Manhattan land ar ea (mile^2) 1.64 input 23 CellSubs in low er Manhattan cellsite 412,273 calc 6 Low er Manhattan land area (mile^2) 1.72 input 24 Market Share 10% input 7 Nightime PopDensity pops/mile^2 52,378 calc 25 Own subs in lower Manhattan cellsite 41,227 calc 8 Nightime Pops in low er Manhattan 90,090 calc 9 Nighttime Pops in midtow n Manhattan 85,900 calc 10 Pop Inc rease btw n Night to Day 1,901,764 calc Voice Traffic Load per Subscriber 11 Pop Increase into lower due to migration into Manhattan 973,522 calc 26 Voice Minutes of Use/year/Sub 1,700 input 12 Pop Increas e into midtow n due to migration into Manhattan 928,242 calc 27 Minutes of Use/w ork-day/Sub 6.54 calc 28 Busy hour concentration of minutes 20% input 13 Pop Incr eas e into low er due to 25% migration w ithin Manhattan 190,369 calc 29 Busy Hour Minutes/Sub during work day 1.31 calc 14 Pop Incr ease into midtow n due to 25% migration w ithin Manhattan 181,515 calc 30 Busy Hour voice traffic load/Sub (mE) 21.8 calc 15 Total day time Pops in low er Manhattan 1,253,981 calc 16 Total day time Pops in midtow n Manhattan 1,195,657 calc 17 Total Day time PopDensity in low er (mile^2) 729,059 calc 18 Total Day time PopDensity in midtow n (mile^2) 729,059 calc

Data Traffic Load per Subscriber Dat a Only CellSite Capacit y f or CDM A2000 EVDO Re v A all IP 31 Average Wired BB Downstream Busy Hour Data Rate/Sub (kb/s) 25 input 37 Dow nstream throughput for 3Sector Site (Mb/s) 2.7 input 32 Ratio of Upstream to Dow nstream Data Rate 20% input 38 Upstream thr oughput for 3Sector Site (Mb/s ) 1.35 input 33 Throttle on Wireless Data Rate Ratio to Wired Rate 4.13% input 39 Dow nstream Capacity/sub requirement for data traffic (kb/s) 1.03 calc 34 Busy Hour Concentration Factor (percent of all daily usage) 20% input 40 Ups tream Capacity/s ub requirement for data traffic (kb/s ) 0.21 calc 35 Percent of Subs who Take Data Service 10% calc 41 Dow ns tream Capacity/s ite requirement f or data traf fic (Mb/s ) 4 calc 36 Equivalent dow nstream megabytes/sub/month of Data Usage 50.0 calc 42 Ups tream Capacity/s ite requirement f or data traf fic (Mb/s ) 1 calc 43 Number of Channels Required for Dow nstream 2 calc 44 Number of Channels Required for Upstream 1 calc 45 Worst Case Number of Channels Required 2 calc

Voice Only CellSite Capacity for CDMA2000 EVDO Rev A all IP Computation of Spectrum Required 46 Dow nstream thr oughput for 3Sec tor Site (Mb/s ) 2.7 input 52 Number of 1.25MHz channelc hannel pairs requiredrequir ed for VoiceV oic e 10 calc 47 Upstream thr oughput for 3Sec tor Site (Mb/s ) 1.35 input 53 Required Spectrum for Voice (MHz) 25 calc 48 Capacity requirement for voice traffic (kb/s) 10 input 54 Number of 1.25MHz channel pairs required for Data 2 calc 49 Voice Circuits Available (upstream limited) 135 calc 55 Required Spectrum for Data (MHz) 5 calc 50 Traf fic Capacity/3Sector Site in Erlangs @2% Blocking 87.97 calc 56 Total Spectrum Required for Both Data+Voice (MHz) 30 calc 51 Number of Voice Subs Supported/Site/Channel 4,036 calc 57 Ratio of Data Channels/Total Channels 17% calc

• The base population in the CBD is Figure 6 Lower and Midtown Manhattan Central the nighttime population (lines Business Districts 7&8)

• One component of increase above the nighttime base population is migration into Manhattan from elsewhere. The computations in the table assume this migration is divided between lower and midtown based on their respective land areas (lines 11&12)

• A second component of increase in the CBD is based upon an estimated 25% migration from Lines 1 through 18 of the Table 5 other parts of Manhattan into the model shows the process of arriving at a CBD (lines 13&14) 2 population density of 729,059 pops/mi which is the same in both Lower and Midtown • The resulting work week daytime districts. population density in the CBD is computed and is the same for both The approach presented in the table lower and midtown districts (lines for estimating the worst case NYC population 17&18) density in the central business districts (CBD)is based upon: CellSite Loading Model Flow Chart

Although the standards for modern 3G The total spectrum required to support air interfaces support smaller cell sizes, we today's voice+modest data load is computed assume for this analysis the smallest to be 30MHz in line 56 of the Table 5 model economic macrocellular architecture is built for a 10% market share carrier but would need using 0.5 mile radius 3 sector macrocells as to be on the order of 75MHz for the mature shown in Figure 7. carriers who enjoy 25% market share. The process for computing the required spectrum is shown in the Figure 8 flow chart. Figure 7 Conventional Macrocellular Architecture

Figure 8 Flow Chart for Macrocellular Analysis Compute Number of Busy Hour Subs in Estimate Voice Traffic Estimate Data Traffic 0.5Mile Worst Ca se Macrocell in Most Dense NYC Load on Worst Case Load on Worst Case Macrocells Busine ss District Macrocell Macrocell Research Manhattan Day Research per Subscriber Research per Subscriber vs Night Pops and Land Current Broadband Wired Current Annual Minutes of Area in Lower and Midtown UP/Downs tream Data Use (MOUs) Manhattan Consumption per Month

Scale Wired Data Compute Subs in Worst Translate MOUs to Busy Consumption to a Smaller Case Lower or Midtown Hour Traffic Load per Fraction for use as Estimate Business District Subscriber for Wireless Data Load

As s ume Subs criber Using 2% GOS Employ Compute Subs in Most Adoption of Data as a Traffic Theory to Compute Smaller Percentage of Total Dense Macrocell within the Effective Mb/s Cellsite Voice Subscriber to District Load/Sub Compute Mb/s Cellsite Load/Sub

Compute Total Voice+Data Load on Cellsite for All Subs in The Macrocell Under Study

Compute Number of Capacity Limits of 1.25MHz Channel Pairs CDMA2000 EVDO-Rev A 3 Needed to Accommodate Sector Cellsite in Up and the Worst Case Downstream Max Up/Downstream Traffic Throughput/Channel Pair in Employing the CDB population Load Mb/s

Compute Total density computed above, the number of Up/Downstream Spectrum subscribers in the worst case 0.5 mile cellsite Needed in MHz is computed in lines 19 through 25 of the Table 5 model. The analysis assumes the following key inputs to the model: The cellsite area in line 20 is multiplied by the CBD population density to arrive at line 21 • Each macrocell employs CDMA2000 pops in the macrocell. Using the CTIA 72% EVDO-Rev A all IP technology figure for cellular adoption results in the cellular subscribers in the macrocell in line • The traffic load is typical of today's 23. On line 24 the market share for the cable cellular system averages of company is assumed to be 10% comparable to that enjoyed by t-Mobile but much less than – Voice traffic of 142 the 25% market share enjoyed by Verizon and MOUs/month AT&T/Cingular. This results in a line 25 estimate of 41,227 of your own subscribers – Data traffic based on 10% generating traffic in your worst case fraction of voice subscribers macrocell. who consume 50 megabytes/month of downstream data

• The market share of the carrier under study is 10% Data Only Cellsite Capacity Voice Traffic Load Lines 37 though 45 of Table 5 Lines 26 through 30 of the Table 5 compute the number of channels and spectrum model compute the voice traffic load. required to support the upstream and Starting with the CTIA reported average downstream data only traffic loads on the 1,700 minutes of use per year per subscriber CBD cellsite during the busy hour. Line 37 (MOUs) of voice traffic, the number of MOUs represents the downstream throughput per site per work day is computed on line 27 and for a three sector EVDO-Rev A macrocell based upon an assumption of 20% of all daily while line 38 represents the upstream minutes being used in the busy hour, the busy throughput limit for the site. These hour MOUs is computed on line 29 and throughputs are based on a private shown in mili-Erlangs per subscriber during communications to me of the results of a the busy hour of 21.8 mE on line 30. simulation conducted by Nortel of various 3G and 4G air interfaces under mobility Data Traffic Load conditions.

Lines 31 through 36 of the Table 5 The total traffic load per subscriber is model are employed to compute the busy hour computed by using lines 31 through 36 of the non-voice data traffic load. Line 31 model as input to compute line 39 and line 40 represents today's average daily downstream of the model. By using the number of data consumption of today's wired cable modem subscribers the total down and upstream load service as reported by a cable operator. Since on the site is computed on lines 41 and 42. modern air interfaces generally support higher Dividing line 31 by line 37 for downstream downstream (i.e. base to mobile) throughput and line 42 by line 38 for upstream yields the then upstream (i.e. mobile to base), the number of 1.25MHz channel pairs needed to upstream subscriber data consumption is support the non-voice data traffic load on computed based on a ratio of upstream to lines 43 and 44 and the worst case of down downstream of 20%. and upstream resulting need for channel pairs is shown on line 45. Line 33 of the model represents a factor to scale the wireless rate to a 4.13% Voice Only Cellsite Capacity fraction of the wired rate. The 4.13% number was computed in order to generate a typical Lines 46 through line 51 of the Table data consumption of 50 megabytes per data 5 model compute the maximum number of subscriber per month (MB/mo) as shown on voice subscribers that a cellsite can support in line 36. Further more, line 34 estimates the order to maintain a 2% grade of service busy hour usage by assuming that 20% of all blocking factor. The throughput on line 47 daily usage occurs during the busy hour. and the assumption that a header suppressed Finally line 35 assumes that data subscribers VoIP channel will require 10 kilobits per represent only 10% of total voice subscribers second on the channel computes to a line 49 (i.e. data consumption averaged across all result of 135 voice circuits supported on the voice subscribers is 5 MB/mo). site. The Erlang B model is employed to compute the maximum traffic capacity of the site in Erlangs and by dividing the voice load that the macrocell network continues to exist in mE per subscriber from line 30, the line 51 to support high mobility applications as well result of a maximum load of 4,036 voice as allow for less than 100% coverage by a subscribers per site is computed. microcell network that underlies the macrocells. Two microcell underlay options Computation of Spectrum Required were considered in this study: 1) Borrowing licensed spectrum from the macrocell pool of The data only capacity from line 45 spectrum to use in microcells and 2) Using and the subscriber voice limit from line 51 is unlicensed WiFi spectrum for the microcells. employed in lines 52 to 57 of the Table 5 In this section we analyze the licensed model to compute the line 56 result of 30MHz spectrum case. of total spectrum required. Examination of the technical On line 52 the number of channel pairs specification for CDMA2000 handsets reveals for voice is computed by simple division of that compliant devices have the ability to line 25 voice subs in the site by the capacity backoff their transmitter power to support per channel on line 51. operation cell sizes much smaller than the 0.5 mile macrocells assumed in the previous On line 54 the number of channel pairs analysis. In Table 6 we compute the range to for data is repeated from line 45. By be limited to about 350 feet under the case of multiplication of the channel pairs needed on maximum power backoff. Allowing for some line 53 for voice and line 54 for data by overlap at the edges, this supports microcell 2.5MHz of spectrum per channel pair the line sizes as small as a 300 feet coverage radius. 53 resulting spectrum need is computed for voice and the line 55 corresponding spectrum need for data is computed. Line 53 and line Table 6 Coverage Range Computation of CDMA2000 Compliant Cellular Handsets 54 are added together to result in the need for 30MHz of spectrum shown on line 56 of the Receiver Sensitivity -104 Table 5 model. (dBmW) Transmitter Minimum Backoff Power -25 (dBmW) As a perspective on the relative drivers Link Budget 79 of bandwidth, the ratio between data and total (dB) Free Space Propagation Range @ 2000MHz 350 channels is computed as 17% on line 57. (feet)

Free Space LinkBudget(dB)=36.56+20Log10(Frequency in MHz)+20Log10(Distance in Miles) Source: “Recommended Minimum Performance Standards for CDMA2000 Spread Spectrum MICOCELLULAR ANALYSIS WITH Mobile Stations”, Release B, Version 1, 3GPP2C.S0011-B, December 13, 2002 LICENSED SPECTRUM

Unlike competitive carriers, cable With reference to Figure 9, it can be operators have extensive HFC plant seen that a 0.5mile macrocell can be filled throughout their franchise area that can be with 77 microcells of 300 foot microcells and used to provide economic backbone using the ability of CDMA2000 to allow 1:1 interconnection between microcells with frequency reuse, the same spectrum can be coverage range much less than 0.5 mile used in each cellsite. Assuming the limited radius. For the microcell case, it is assumed space to mount microcells and the need to keep the cost low, it is assumed that an omnidirectional antenna is used (i.e. single sector) and that a maximum of one channel pair per microcell is supported. The figure Figure 10 Flow Chart for Microcellular Analysis also shows the application of cable plant as a Compute Number of Busy Hour Subs in Compute Voice Traffic Compute Data Traffic Worst Ca se Microcell in Most Dense NYC Load on Worst Case Load on Worst Case Busine ss District Microcell Microcell backbone for connection of the cells. A Research Manhattan Assume for a Test Case Day vs Night Pops and a Number for Subscriber Land Area in Lower and Current Annual Minutes variety of backbone technologies are possible Midtown Manhattan of Use (MOUs) including DOCSIS to stand alone Assume for a Test Case Compute Subs in Worst Translate MOUs to Busy a Number For Case Lower or Midtown Hour Traffic Load per MB/Month/Subscriber microcellular base stations as well as radio Business District Subscriber over fiber to connect a single base station to Wireless Data Load Assume Subscriber Using 2% GOS Employ Adoption of Data as a Compute Subs in Most remote antenna drivers within each microcell. Traffic Theory to Smaller Percentage of Dense Microcell within Compute Effective Mb/s Total Voice Subscriber to the District Cellsite Load/Sub Compute Mb/s Cellsite Load/Sub

Compute Total Voice+Data Load for All Subs in The Microcell Under Study Figure 9 Microcell Architecture Assuming a Single Capacity Limits of 1.25MHz Channel Pair CDMA2000 EVDO-Rev A Single Omni Sector Compute the Percent Up and Downstream Loading on Microcell for Max the Offered Traffic for Throughput/Channel Pair Each Test Case in Mb/s

0.5Mile CablePlant Macrocells Microcell Backbone The analysis considers several cases in 77 300’ which the inputs to the model are varied: Microcells

• Each microcell employs CDMA2000 EVDO-Rev A all IP technology with CDMA2000 Base single channel pair and an omni-sector Station antenna

Model Flow Chart • The market share of the carrier under study is 10% Unlike the approach employed for computing the spectrum required per • The traffic loads studied consist of two macrocell based on traffic, in this licensed cases microcellular option under study it is assumed there is only one channel pair allocated to • Today's voice with multimedia data each microcell. The analysis approach focus traffic max'ed out to use full capacity of the microcell is upon understanding the capacity of such an architecture under several traffic load cases – Voice of 142 MOUs and then to compare the traffic capacity of this microcellular+macrocellular hybrid to the – Each data sub consumes previous macrocellular only case. 653MB/month downstream traffic

The process for computing the – Data subs are 10% of voice capacity of the microcellular architecture is subs shown in the Figure 10 flow chart. • No voice traffic with data capacity max'ed out

– All voice traffic on the macrocell and none on the microcells

– Each data sub consumes capacity to handle today's voice plus non- 816MB/month downstream voice data load. So the analysis objective in traffic this case is not to compute how much

– Data subs are 10% of voice spectrum is needed, but instead to compute subs how much more than today's traffic can the microcell support.

Table 7 Licensed Spectrum Micro Cell Analysis Model for CDB in Manhattan

Manhattan Demographics CellSite Loading 1 Manhattan land area (miles^2) 28.4 input 19 CellRadius (miles) 0.06 input 2 Nighttime population 1,487,536 input 20 CellArea (miles^2) 0.01 calc 3 Daytime Population 3,389,300 input 21 Daytime Pops in lower Manhattan cellsite 7,394 calc 4 Ratio Day/Night 2.3 calc 22 Cellular Adoption 72% Input 5 Mid-town Manhattan land area (mile^2) 1.64 input 23 CellSubs in lower Manhattan cellsite 5,324 calc 6 Lower Manhattan land area (mile^2) 1.72 input 24 Market Share 10% input 7 Nightime PopDensity pops/mile^2 52,378 calc 25 Own subs in lower Manhattan cellsite 532 calc 8 Nightime Pops in lower Manhattan 90,090 calc 26 Voice Minutes of Use/year/Sub 1,700 input 9 Nighttime Pops in midtown Manhattan 85,900 calc 27 Minutes of Use/work-day/Sub 6.54 calc 10 Pop Increase btwn Night to Day 1,901,764 calc 28 Busy hour concentration of minutes 20% input 11 Pop Increase into lower due to migration into Manhattan 973,522 calc 29 Busy Hour Minutes/Sub during work day 1.31 calc 12 Pop Increase into midtown due to migration into Manhattan 928,242 calc 30 Busy Hour voice traffic load/Sub (mE) 21.8 calc 13 Pop Increase into lower due to 25% migration within Manhattan 190,369 calc 31 Voice Load on Site from own subs (Erlangs) 11.6 calc 14 Pop Increase into midtown due to 25% migration within Manhattan 181,515 calc 32 Number of Voice Ckts to be Reserved for 2% QoS Blocking 18 calc 15 Total daytime Pops in lower Manhattan 1,253,981 calc 33 Capacity requirement for voice traffic (kb/s) 10 input 16 Total daytime Pops in midtown Manhattan 1,195,657 calc 34 Upstream & Downstream l oad on Si te from Voice Ckts i n Mb/s 0.180 cal c 17 Total Daytime PopDensity in lower (mile^2) 729,059 calc 18 Total Daytime PopDensity in midtown (mile^2) 729,059 calc

Data Traffic Load per Subscriber MicroCellSite Capacity for Single Channel CDMA2000 EVDO Rev A all IP 35 Average Wired BB Downstream Busy Hour Data Rate/Sub (kb/s) 25 input 41 Downstream throughput for Omni Single Sector Site (Mb/s) 0.9 input 36 Ratio of Upstream to Downstream Data Rate 20% input 42 Upstream throughput for Omni Single Sector Site (Mb/s) 0.45 input 37 Throttle on Wireless Data Rate Ratio to Wired Rate 54.00% input 43 Downstream Site throughput left after voice reserve (Mb/s) 0.720 calc 38 Busy Hour Concentration Factor (percent of all daily usage) 20% input 44 Upstream Site throughput left after voice reserve (Mb/s) 0.270 calc 39 Percent of Subs who Take Data Service 10% input 45 Downstream Capacity/sub requirement for data traffic (kb/s) 13.50 calc 40 Equivalent downstream megabytes/sub/month of Data Usage 653.1 calc 46 Upstream Capacity/sub requirement for data traffic (kb/s) 2.70 calc 47 Total Downstream data traffic Load/Site (Mb/s) 0.719 calc 48 Total Upstream data traffic Load/Site (Mb/s) 0.144 calc 49 Total Downstream Data+Voice Traffic Load on Site (Mb/s) 0.899 calc 50 Total Upstream Data+Voice Traffic Load on Site (Mb/s) 0.324 calc

Ratio Analysis 51 Ratio of DS Voice Load/Site to Total Single Channel Site Capacity 20% calc 52 Ratio of DS Data Load/Channel to Total Single Channel Site Capacity 80% calc 53 Ratio of DS Data Load/Channel to Max Possible Data Load 100% calc 54 Ratio of US Voice Load/Site to Total Single Channel Site Capacity 40% calc 55 Ratio of US Data Load/Channel to Total Single Channel Site Capacity 32% calc 56 Ratio of US Data Load/Channel to Max Possible Data Load 53% calc

Licensed Microcellular Capacity Computation Starting at line 31 in the Table 7 model the Erlang voice load on the microcell In a matter parallel to the is computed and using Erlang theory the macrocellular analysis model, the model of number of voice circuits that must be reserved Table 7 was created to compute the capacity to support a 2% grade of service blocking is of the licensed microcellular alternative based computed on line 32. Using the voice data upon a single CDMA2000 channel pair in a rate on line 33, the load on the microcell in 300 foot omni-sector microcell using an Mb/s needed to support today's voice traffic is EVDO-Rev A air interface. computed on line 34.

The Table 7 model tracks the Table 5 Lines 35 through 40 in the Table 7 macrocell model from lines 1 through line 30. model directly parallel the computation in The analysis departs from the macrocell case macrocell model. The input parameters for beyond line 30, because even this single the data consumption are adjusted upward so channel pair microcell has more than enough that the capacity of the microcell is max'ed out. Line 37 of the model is set so that line 52 300 foot radius of coverage comparable to the of the model just reaches 80% of site capacity same coverage radius assumed for the which when added to the 20% needed for licensed microcellular option considered voice results in 100% of the site capacity above. consumed by the offered traffic. Table 8 Typical Performance of WiFi Network The finding from this analysis is Access Points Protocol Release Operating Frequency Data Rate Data Rate Range Range shown on line 40 of the Table 7 model Date (T yp) (Max) (Indoor) (Outdoor) 802.11a 1999 5.15-5.35 / 5.47-5.725 / 25 Mbit/s 54 Mbit/s ~25 ~75 indicating that a licensed microcellular 5.725-5.875 GHz meters meters 802.11b 1999 2.4-2.5 GHz 6.5 Mbit/s 11 Mbit/s ~35 ~100 meters meters network can support all of today's voice 802.11g 2003 2.4-2.5 GHz 25 Mbit/s 54 Mbit/s ~25 ~75 meters meters traffic as well as allowing non-voice data 802.11n 2007 2.4 GHz or 5 GHz bands 200 Mbit/s 540 Mbit/s ~50 ~125 draft meters meters usage to grow from the 50MB/month/data-sub Source: Wikipedia entry for 802.11 to 653MB/month/data-sub. What is more remarkable is that unlike the macrocellular Unlike the approach employed for case which required 30MHz of spectrum to computing the spectrum required per support the much smaller non-voice data load, macrocell based on traffic, in this unlicensed with the microcellular architecture only a microcellular option under study it is assumed single channel pair of 2.5MHz (i.e. 2 x there are a fixed number of four WiFi 1.25MHz) is needed. channels allocated to each microcell. As with the licensed microcellular option, this analysis The licensed microcell model of Table approach focus is upon understanding the 7 was also run with the voice traffic set to capacity of such an architecture under several zero such as would be the case if all voice traffic load cases and then to compare the traffic was supported on the macrocell overlay traffic capacity of this and the microcellular network employed only microcellular+macrocellular hybrid to the for data. Under these conditions, the model previous macrocellular only case. computes a non-voice data only capacity that supports each data sub consuming Model Flow Chart 816MB/month of downstream non-voice data traffic. The process for computing the capacity of the unlicensed microcellular MICOCELLULAR ANALYSIS WITH architecture is shown in the Figure 11 flow UNLICENSED SPECTRUM chart.

This alternative differs from the just Figure 11 Flow Chart for Unlicensed Microcellular licensed microcell analysis, just above, in that Analysis

Compute Number of Busy Hour Subs in Compute Voice Traffic Compute Data Traffic Worst Ca se Microcell in Most Dense NYC Load on Worst Case WiFi Load on Worst Case WiFi unlicensed WiFi spectrum is employed in the Busine ss District Microcell Microcell Research Manhattan Assume for a Test Case Day vs Night Pops and a Number for Subscriber microcells in an architecture similar to today's Land Area in Lower and Current Annual Minutes muni-WiFi systems. As with the previous Midtown Manhattan of Use (MOUs) Assume for a Test Case Compute Subs in Worst Translate MOUs to Busy a Number For Case Lower or Midtown Hour Traffic Load per MB/Month/Subscriber licensed microcell analysis, the macrocell Business District Subscriber Wireless Data Load network continues to overlay the unlicensed Assume Subscriber Using 2% GOS Employ Adoption of Data as a Compute Subs in Most Traffic Theory to Smaller Percentage of Dense Microcell within Compute Effective Mb/s Total Voice Subscriber to microcells to support high mobility and fill the District Cellsite Load/Sub Compute Mb/s Cellsite any microcell coverage gaps. Load/Sub Compute Total Voice+Data Load for All Subs in The WiFi Microcell Under Study

Capacity Limit of 4chan Compute the Percent 802.11b Omni Sector Loading on Microcell for Up and Downstream the Offered Traffic for Max Each Test Case Examination of the performance of Throughput/microcell site WiFi network access points in Table 8 shows that such outdoor unlicensed microcells have Table 9 Unlicensed Spectrum Micro Cell Analysis Model for CDB in Manhattan

Manhattan Demographics CellSite Loading 1 Manhattan land area (miles^2) 28.4 input 19 CellRadius (miles) 0.06 input 2 Nighttime population 1,487,536 input 20 CellArea (miles^2) 0.01 calc 3 Daytime Population 3,389,300 input 21 Daytime Pops in lower Manhattan cellsite 7,394 calc 4 Ratio Day/Night 2.3 calc 22 Cellular Adoption 72% Input 5 Mid-town Manhattan land area (mile^2) 1.64 input 23 CellSubs in lower Manhattan cellsite 5,324 calc 6 Lower Manhattan land area (mile^2) 1.72 input 24 Market Share 10% input 7 Nightime PopDensity pops/mile^2 52,378 calc 25 Own subs in lower Manhattan cellsite 532 calc 8 Nightime Pops in lower Manhattan 90,090 calc 26 Voice Minutes of Use/year/Sub 1,700 input 9 Nighttime Pops in midtown Manhattan 85,900 calc 27 Minutes of Use/work-day/Sub 6.54 calc 10 Pop Increase btwn Night to Day 1,901,764 calc 28 Busy hour concentration of minutes 20% input 11 Pop Increase into lower due to migration into Manhattan 973,522 calc 29 Busy Hour Minutes/Sub during work day 1.31 calc 12 Pop Increase into midtown due to migration into Manhattan 928,242 calc 30 Busy Hour voice traffic load/Sub (mE) 21.8 calc 13 Pop Increase into lower due to 25% migration within Manhattan 190,369 calc 31 Voice Load on Site from own subs (Erlangs) 11.6 calc 14 Pop Increase into midtown due to 25% migration within Manhattan 181,515 calc 32 Number of Voice Ckts to be Reserved for 2% QoS Blocking 18 calc 15 Total daytime Pops in lower Manhattan 1,253,981 calc 33 Capacity requirement for voice traffic (kb/s) 10 input 16 Total daytime Pops in midtown Manhattan 1,195,657 calc 34 Upstream & Downstream load on Site from Voice Ckts in Mb/s 0.180 calc 17 Total Daytime PopDensity in lower (mile^2) 729,059 calc 18 Total Daytime PopDensity in midtown (mile^2) 729,059 calc

Data Traffic Load per Subscriber MicroCellSite Capacity for Single Channel CDMA2000 EVDO Rev A all IP 35 Average Wired BB Downstream Busy Hour Data Rate/Sub (kb/s) 25 input 41 Typical Down+Upstream throughput per 802.11b channel (Mb/s) 6.5 input 36 Ratio of Upstream to Downstream Data Rate 20% input 42 Number of active 802.11b channels/access point 4 input 37 ThrottleT hrottle on WirelessWirel ess Data Rate Ratio to Wired Rate 112.00% input 43 Derating percentage based on contention 70% input 38 Busy Hour Concentration Factor (percent of all daily usage) 20% input 44 Down+Upstream throughput for Omni Single Sector Site (Mb/s) 18.2 calc 39 Percent of Subs who Take Data Service 100% input 45 Down+Upstream Site throughput left after voice reserve (Mb/s) 17.84 calc 40 Equivalent downstream megabytes/sub/month of Data Usage 1355 calc 46 Downstream Capacity/sub requirement for data traffic (kb/s) 28.00 calc 47 Upstream Capacity/sub requirement for data traffic (kb/s) 5.60 calc 48 Total Down+Upstream data traffic Load/Site (Mb/s) 17.888 calc 49 Total Downstream Data+Voice Traffic Load on Site (Mb/s) 18.068 calc

Ratio Analysis 50 Ratio of Voice Load/Site to Total WiFi Site Capacity 1% calc 51 Ratio of Data Load/Channel to Total WiFi Site Capacity 98% calc 52 Ratio of DataLoad Max Possible Data Load 100% calc

The analysis considers several cases in – Data subs are 100% of voice which the inputs to the model are varied: subs

• Each microcell employs 4 channels (1, — No voice traffic with data capacity 4, 8,11) of 2400MHz 802.11b WiFi max'ed out network access points. – All voice traffic on the • The market share of the carrier under macrocell and none on the study is 10% microcells

• The traffic loads studied consist of two – Each data sub consumes cases however the capacity of the 1,379MB/month downstream system is now so great, we assume the traffic limit case in which non-voice data adoption grows from the 10% – Data subs are 100% of voice assumption employed in the licensed subs cases to approach 100% adoption of non-voice data services by basic voice Unlicensed Microcellular Capacity subscribers. Computation

— Today's voice with multimedia In a matter parallel to the licensed data traffic max'ed out to use full microcellular analysis model, the model of capacity of the microcell Table 9 was created to compute the capacity – Voice of 142 MOUs of the unlicensed microcellular alternative based on a muni-WiFi type of architecture – Each data sub consumes employing 802.11b technology. 1,355MB/month downstream traffic The Table 9 model is the same as the

licensed microcellular model from line 1 through line 40 with the exception that the multimedia services because the wide area line 37 data load is set to max out the cellsite microcell is always available for fallback capacity as shown in line 52. Lines 41 when microcell coverage may be lacking or through line 43 have been added to unlicensed marginal. model to compute the cellsite (i.e. 4 channel WiFi access point) capacity on line 44. Also of interest is designing non-voice Otherwise this model is the same as that used applications which take advantage of the for the licensed microcellular case. increasingly low cost but very capable storage and intelligence in today's cellphones to Line 41 of the Table 9 model shows an implement applications in which low latency assumption for a typical data throughput of low bandwidth user interface traffic is 6.5Mb/s per WiFi 802.11b channel which is delivered over the wide area macrocell and multiplied by 4 channels on line 42 and high latency tolerant block file transfers are further by a 70% derating factor on line 43 to conducted using the microcells. Such an result in the throughput of 18.2Mb/s on line approach in which the wide area network is 44. The 70% factor is meant to represent employed for user interface and control and overhead associated with the WiFi contention the very wide bandwidth delay insensitive protocol. Employing more recent 802.11X traffic can be delivered over the microcellular technology could increase the throughput. data network without any noticeable impairments notice by user for many Because the voice load on the site is so applications of interest, e.g: small in comparison to the non-voice data load, there is virtually no difference in the • Synchronizing mp3 music files stored maximum non-voice capacity of site with or locally in the phone with an online without voice traffic. The model computes a server maximum capacity of 1,355MB/month/sub with voice traffic on the microcell and • Enhancing Cache and Carry of video 1,379MB/month/sub with voice traffic moved by supporting intervals of a few to the macrocellular overlay network. minutes between reloading the cache

Since today's cable modem subscriber • Automatically delaying large email consumes 1,209MB/month/subscriber of file uploads or downloads until within datastream high speed internet services, this range of a microcell unlicensed microcellular network has a capacity that exceeds today's wired SUMMARY AND CONCLUSIONS cablemodem data consumption. Although cable operators may ALTERNATIVE ARCHITECTURES currently own 20MHz of spectrum versus competitors holding 2-3 times more, cable can One concern that might be raised with more than compensate by exploiting unique regard to implementation of a microcellular backbone cableplant assets that would be non- network is the difficulty of obtaining reliable economic for competitors to match. Stealing 100% area coverage with such small 300 foot one 1.25MHz channel pair from the macrocell radius cells. However, this concern is inventory to implement a single channel pair mitigated when you consider that the underlay microcell underlay would support today’s need not have 100% coverage of the voice traffic load as well as allowing data macrocell in order to support high quality subscribers to scale from today’s 10% adoption to 100% adoption at the same an unimpaired level of user satisfaction 50MB/month/sub rate of consumption. Using resulting in significant CapEx cost savings for four unlicensed 802.11b channels in the the microcell underlay. microcell in a MuniWiFi like architecture would allow cable to offer competitive wide By intelligent design of the services area voice services as well as support 100% offered, the coverage quality of the microcell adoption by subs who consume data at rates underlay need not be made equal to the even greater than today’s wired cablemodem macrocell overlay network in order to provide subscribers.

A summary of the modeling results in Table 10 demonstrates the significant increase of a microcellular network versus conventional macrocellular plant.

Table 10 Summary of Systems Capacity Modeling Results

Portion of Licensed Total Portion of UnLicensed Per Subs cr iber Tr affic Load Total Licensed Spectrum Needed UnLicensed Spectrum Needed Case Under Study Ar chite cture Voice Tr affic Dow nstre am Data Spectrum For Voice For Data Spectrum For Voice For Data Load Consumption Employed Traffic Traffic Employed Traffic Traffic (Minutes of Use (MegaBytes per (M Hz) (%) (%) (M Hz) (%) (%) Per Month) Month) Macrocells with Unlimited 0.5Mile radius 3-sector 142 5 30=1.25X2X12 83% 17% None 0% 0% Licensed Spectrum EVDO-Rev A all IP DataSubs=10% CDMA2000 Each Sub 50MB Underlay of Microcells 300' radius single 142 5 2.5=1.25X2X1 40% 2% None 0% 0% Employing a Single channel omni sector DataSubs=10% Channel Pair EV DO- Rev A all IP Each Sub 50MB Mix ed V oic e+Data CDMA2000 Underlay of Microcells 300' radius single 142 65 2.5=1.25X2X1 40% 60% None 0% 0% Employing a Single channel omni sector DataSubs=10% Channel Pair EV DO- Rev A all IP Each Sub=653MB Voice+Maxed Out Data CDMA2000 Underlay of Microcells 300' radius single 142 65 2.5=1.25X2X1 40% 60% None 0% 0% Employing a Single channel omni sector DataSubs=100% Channel Pair EV DO- Rev A all IP Eac h Sub=65 MB Voice+Maxed Out Data CDMA2000 Underlay of Microcells 300' radius single 0822.5=1.25X2X1 0% 100% None 0% 0% Employing a Single channel omni sector DataSubs=10% Channel Pair EV DO- Rev A all IP Each Sub=816MB Max ed Out f or Data Only CDMA2000 Underlay of WiFi 300' radius four 0 1209 None 0% 0% 74MHz 0% 88% Microcells channel omni sector DataSubs=100% 4 Chans Today's Wired Data 802.11b WiFi Each Sub=1209MB 1, 4, 8, 11 Consumption Underlay of WiFi 300' radius four 0 1379 None 0% 0% 74MHz 0% 100% Microcells channel omni sector DataSubs=100% 4 Chans Max ed Out f or Data Only 802.11b WiFi Each Sub=1379MB 1, 4, 8, 11

EXTENDING CONDITIONAL ACCESS SYSTEMS TO SUPPORT DRM SYSTEMS Balu Ambady Cable Television Laboratories, Inc. Bill Helms Time Warner Cable, Inc.

Abstract content may be considered to be moving between one of the three content protection Conventional implementations of content domains as shown in Figure-1: security by Multiple System Operators (MSOs) rely on hardware based broadcast Conditional Access Systems (CAS). However, competitors like IPTV providers, and Internet based movie-on-demand sites have started offering high value content using content protection based on software Digital Rights Management (DRM) systems like Microsoft WMDRM, and Real Helix DRM. It Figure 1 – Content Protection Domains appears that content providers are increasingly becoming comfortable with the 1) Authorized Service Domain (ASD) reduced level of security provided by DRM systems for certain levels of content. Within the ASD, content is secured using the mechanisms provided by the MSO Under currently available solutions, according to usage rules set by the while the content is under CAS protection, Operator’s provisioning system. Content is opportunities to offer innovative content protected using CAS while it is transferred packages to the customer are preserved. from the MSO head-end to the subscriber home, and by an ASD specified protection This paper describes architectural options mechanism while it moves between one or where CAS protected content can be more trusted devices that are part of the ASD transferred securely from Set Top Box (STB) network. devices to Portable Media (PMD) or PC devices with DRM protection. This would 2) Approved Output Domain (AOD) allow the operator to provide an enhanced home networking and content usage As permitted by FCC encoding rules1, experience to the subscriber. content may also be released to non-operator Content Protection (CP) or DRM systems that have been approved2 by CableLabs®. INTRODUCTION When the content is released from the ASD to another CP system, it is considered to exit Content Protection Domains the ASD and enter the AOD. The content is still securely protected by the CP/DRM When user subscribed content is moving system and the encoding rules are enforced from the Operator’s head-end device to a by the CP/DRM. STB, and then on to a Home Network device, from a content protection point of view, the Several Link Protection systems like • Interoperability with other devices in DTCP and HDCP, and DVD based recording the home solutions like VCPS fall under the AOD category. A PC based architecture, the 2) Content Provider/owner OCUR (OpenCable Unidirectional Receiver) using the Microsoft WMDRM and Real • Sufficient protection of owner rights Helix DRM systems for content protection also fall under this category. • Monetization of content

3) Unprotected Output Domain (UOD) 3) MSO

Lower value content that require only • Ensuring a link between services minimal levels of protection, or no protection provided and payment received at all may be released to the UOD, for example to the unprotected Analog video • Support for flexible business models, outputs. freedom to innovate

High Level Design Goals • Leverages existing infrastructure

where possible While a large number of the devices that are capable of serving MSO delivered content fall under one of the three domains • Provide consistent MSO branded described earlier, many new devices like experience PMDs and cell phones do not readily fall under these domains. In order for the MSOs 4) Device manufacturer to offer the best entertainment experience for the subscriber, it is desirable to bring more of • Easy to license technology these new devices either under ASD, or AOD. In order to join the ASD or AOD, • High level of interoperability these devices must be capable of providing sufficient levels of content protection, as well • Freedom to innovate as be capable of supporting the MSO defined usage rules. We offer some architectural ACHITECTURAL OPTIONS options to achieve this goal. We offer four solutions to achieve the The options provided in the next section goals outlined above. Three of the solutions provide solutions while striving to attain a described below use Operator provided balance between the needs of the main security mechanisms and thus allow the stakeholders and their needs: content to stay fully within the ASD, and the fourth option provides a “bridge” mechanism 1) Subscribers from ASD to a third-party DRM solution.

In the case of the three ASD • Ease of use/transparency solutions, encoding rules permit the operator

to extend the usage rules beyond the limited • Fair use usage rules expressed using the 8-bit Copy

Control Information (CCI) bits3. This would allow for new business models for the MSOs, and more choice for the subscriber. For developed by an ASD vendor, and all example, current CCI is not sufficient to vendors follow a standard set of protocols, express a “rent” model where a subscriber content formats and encryption schemes. In may rent a digital content for 7 days. addition to allowing multiple ASD vendors to participate in this eco-system, this has the In the case of the bridge from ASD to the added advantage of allowing the same ASD native DRM space, certain additional infrastructure and standards to be utilized by encoding rule restrictions may apply. both MSO leased and retail devices.

1) Hardware ASD The CA protected content is converted into ASD protected content. Security This architectural approach would require Packages are created that store the usage all devices that are interested in joining the rights to be kept with the encrypted content. ASD to embed the Downloadable The Secure Micro is responsible for key Conditional Access System (DCAS) Secure generation during encryption, and for Micro chip and store the associated keys, retrieval of keys during decryption. The Host secrets and Root of trust. An architectural Transport processor is responsible for the diagram of a typical implementation of home actual encryption/decryption of the content. networked ASD Host with a DVR application is shown in Figure-2. Devices that belong to the same ASD are managed by the ASD controller by using a The System on a Chip design would allow “Trusted List” that identifies the devices that the Root, critical security parameters and are tied to a subscriber billing account. keys to be securely stored on the Secure Microchip. The software modules that support ASD functionality can be securely downloaded on to the device based on this Root of trust. The ASD modules are

Figure 2 – Simplified DCAS ASD Host Block Diagram

2) Cable DRM (CDRM) DRM format and also enforces Licensing rules, and a Content Server from where the An alternate approach would be for the Client can access DRM protected content. Cable industry to develop its own DRM solution that can run on a variety of devices In a simple use case, as the 1st step a that do not have the Secure Microchip or personalized copy of the Cable DRM Client other Hardware secure storage for keys and that is uniquely tied to the PMD is securely other security parameters. Ideally, this would downloaded. The 2nd step, high value content be a Platform/Operating System (OS) that the subscriber is authorized to receive is independent technology that is easy to moved with Conditional Access protection to implement on a wide variety of devices like the Content Server, in this case the STB. The cell phones and PMD made by different 3rd step, as authorized by the License server, manufacturers. Content security will be the STB transcrypts the content in to CDRM based on obfuscation of keys and code, and format. The 4th step, the PMD can download software renewability. this content from the STB and consume it as allowed by the CDRM usage rules. The architecture for such a solution is shown in Figure-3. The components are Since the Cable DRM devices use the similar to typical DRM systems, there is a Operator’s protection system, devices DRM Client on the portable device, a implementing this are part of the ASD. These Controller providing Keying, Registration devices support the extended usage rules and Individualization for the Client, a using standardized protocols as in the case of License Server that packages the content in Hardware ASD solution.

Controller W Billing Keying e 1 b 2 1

License High 2 Value content Server Cable DRM 4 3 Client 4 CDRM protected PMD 4 content 3 Content Server

STB

Figure 3 – Cable DRM

Because of the relative ease of • Use of Protected Media Paths circumventing Software DRM systems, the DRM Client should be carefully designed to o Utilize OS support (e.g., Vista preserve execution integrity and algorithm PVP) secrecy, prevent tampering, impersonation, and input spoofing. o Disk/CD/DVD drive solutions

These security goals are usually achieved • Secure boot up/firmware in conventional software DRM systems by using the following mechanisms: • Encrypted/protected memory

• Use of information diversity and Other design attributes supporting complexity security should include:

• Use of masking techniques to mask: • Secure clock control flow, data/code itself, location, and usage o Anti-rollback time

• Use of one-way transformations, o Secure (external) time source temporal and spatial diversity • Secure random number generation • Use of Index computations, aliasing techniques • Support of standard cryptographic algorithms • Use of “bad programming” practices: e.g., use of pointers, goto • Support for secure Proximity checks

o Hard to debug (on purpose) • Efficient software implementation

o Prevent Dynamic analysis • Robust Revocation, and Renewal (during execution) mechanisms

• Use of encrypted functions o Breach response readiness

o Integrity Verification Kernel o Heart-beat checks

• Use of “white box” techniques • Secure DRM infrastructure

• Resistance to side channel attacks o Secure build servers (“grey-box”) - however, this may be less of a concern as S/N ratios are o Secure (firewalled) download significantly degraded in GP portals • Ability to add features later • Software Key hiding o Many skews to prevent o Watermarking, fingerprinting domino effect (of hacks)

o Ability to use Trusted vendors to port their ASD clients into the Platform Module Portable devices. The architecture and requirements for the Software Client are Similar to the Hardware ASD, the CDRM similar to the CDRM case, except that some would allow the MSOs to use a common of the messaging protocols may be messaging protocol, encryption scheme, data proprietary to the CAS/ASD vendor. Another structures and extended usage rules on all of difference is that instead of a single DRM, the devices on which the CDRM Client is there would be multiple CAS/ASD systems, available. one for each vendor.

3) Software ASD Since the main difference between Hardware ASD and the current solution is This is a modified approach that would that the security is based on obfuscated combine the options-1 and 2 above, this software instead of hardware stored keys, option would allow existing CAS/ASD this

Figure 4 – ASD/DRM Bridge solution maybe considered as a “Software ASD” implementation. Of course, this requires that the bridge device translate from the ASD encryption 4) ASD/DRM Bridge and protocol to the native DRM encryption and protocol. Under this model, the The ASD/DRM Bridge model shown implementation of the bridge to DRM is the in Figure-4 is a compromise between the responsibility of the device vendor, rather previous three models. It uses ASD protocol than the operator or ASD vendor. to exchange content with the ASD Bridge device (e.g., PMD). However, the native Comparison of Options DRM is responsible for securing the content and obeying the MSO usage rules contained in the ASD protocol. The relative advantages and disadvantages References of the four options outlined above are shown in Appendix-A. 1 The FCC Second Report and Order (FCC 03-225) Section VI-B. http://hraunfoss.fcc.gov/edocs_public/attachmatch/FC CONCLUSIONS C-03-225A1.pdf

Depending on the implementation 2 The FCC Second Report and Order (FCC 03-225) timeline, complexity of implementation, ease Section V. http://hraunfoss.fcc.gov/edocs_public/attachmatch/FC of licensing, cost and other factors, the C-03-225A1.pdf Operator may select one of the four options presented above. An ASD solution that 3 SCTE 41 – POD Copy Protection System embraces the new generation of Portable http://www.scte.org/documents/pdf/ANSISCTE41200 devices and other Home Networked devices 4.pdf would help to enhance the subscriber’s entertainment experience and thus would help to strengthen brand loyalty.

Appendix – Comparison of Options

Hardware ASD Cable DRM ASD Client DRM Bridge Complexity of Low Low Low High Licensing Cost to MSO High High Medium Low Complexity of High High Medium Medium Development Interoperability High High Medium Medium Extended usage Yes Yes Yes Yes with possible rules restrictions

HOUSE CHECK - A NEW PROCESS TO VERIFY VOICE INSTALLS Mike Gordish, Product Manager, Mobile Workforce Manager, C-COR Robert F. Cruickshank III, VP OSS Market Strategy, C-COR Dan Rice, VP Assurance Solutions, C-COR

Abstract quality of incumbent switched circuit telephone service. As MSOs accelerate efforts to satisfy strong market demand for cable digital voice Where, initially, operators could afford to services they have an opportunity to adopt devote whatever amount of time it took for new installation procedures that can well-trained in-house installation personnel contribute significantly to customer retention to perform comprehensive testing and repairs while reducing ongoing operational costs. on drop connections and home wiring to Back office-based ‘Assurance’ software that ensure quality performance, the industry is delivers comprehensive analysis of voice long past the stage where spending several service performance on the MTA (multimedia hours per installation is feasible. In order to Terminal Adapter) across all household scale to millions of customers, MSOs have cable outlets with DOCSIS equipment had to reduce installation times while often installed, allows installers to quickly obtain a turning to outside contractors to handle the view of whole-home service quality that volume of orders. simplifies identification and repair of problems at installation and provides a This accelerated pace of deployment and reference base for use in future trouble service provisioning raises the risk that shooting. By aggregating and interpreting problems will arise after installers leave, DOCSIS-generated data on RF signal which in turn can lead to higher levels of quality, jitter, packet loss and other customer dissatisfaction. One MSO’s recent parameters, these “house checks” provide studyi of the impact of faulty service on easy-to-understand readouts on installer customer retention found that 10 percent of handhelds and give Customer Service voluntary disconnects are either directly or Representatives, Dispatchers, Engineers and indirectly caused by service problems. The Network Operations Center (NOC) personnel MSO found that repeat service problems are a reliable record of the true state of the largest single cause behind voluntary performance before the installer leaves the disconnects. premises. Another recent studyii, by an independent INTRODUCTION research company, found that technicians had to visit 34 percent of new cable VoIP Early market response to IP-based cable customers within 90 days of installation and digital voice service has prompted MSOs to that 16 percent required visits two or more move as quickly as possible to extend service times. Twenty-nine percent of the people reach within existing markets and into new experiencing problems said their “primary markets. At the same time, operators frustration” was with recurring quality recognize that the faster they proceed, the deficiencies, while twenty-four percent said greater the challenge they face in meeting the the frustration was with recurring reliability fundamental requirement that their voice problems. services match or exceed the performance With various reports citing truck roll costs of performance parameters it’s easy to of anywhere from $130 to $200 per average ‘fudge’ the results to produce a performance visit, the operational costs combined with the report that makes it seem like a sub-par costs of customer dissatisfaction are clearly service meets minimal threshold levels. If the too high for operators to sustain these levels service is relatively clean and works at the of service problems. The obvious place to time of installation, the fact that levels are begin in the effort to cut these costs is with below threshold may not lead to discernable the installation process itself. Fortunately problems until long after the installer has remedies are at hand that will help operators departed. achieve a much higher level of assurance that service quality is where it should be when the Measurements taken by installers also installer leaves the home without adding provide few clues as to the source of significantly to the amount of time installers problems, with the result that the installer spend executing the work order. may go through time-consuming remedial steps that can be innefectivebefore he knows THE INSTALLATION CHALLENGE he must turn to network technicians for help. And even then he may not know whether the Two major issues stand in the way of an problem is network related or a problem MSO’s ability to ensure digital voice service associated with IP addressing, device is performing at required quality levels upon identification or other data issues that require completion of service installation. The first IT department attention. has to do with the amount of time it takes to thoroughly examine all potential problem Beyond the immediate impediments to areas using traditional means of testing and efficient and successful installation, manual measurement. Getting a good quality read on measurement and data entry provides an levels from the primary outlet to be used for imprecise record of device and link status for connecting the MTA (multimedia terminal use as a reference baseline for adapter) to the cable network is typically all troubleshooting problems in the future. Thus, that time allows, which means that if the current methodologies not only contribute to subscriber subsequently moves the MTA to post-installation problems; they impede another outlet, there is no assurance the same efficient resolution of those problems when performance levels will apply. they occur. Splitter locations, outlet-specific points of ingress, variations in microreflections THE OPTIMUM HOUSE CHECK depending on where lines are capped off can all produce significant variations in Thoroughness in monitoring service performance from one outlet to the next. performance is vital to ensuring potential Knowing where the weak points are gives the sources of future problems are mitigated operator the option of either fixing them or while the installer is still on the job. This advising the customer not to use those includes taking stock of key internal and outlets. external network performance parameters as well as all metrics that are vital to proper Another problem has to do with the fact performance of premise devices. that contract workers are typically paid by the number of jobs they complete per day. Every time they encounter a problem that must be fixed, their time on the job is extended. Unfortunately, with manual entries • Provisioning status with regard to performance of voice and data interfaces, related service elements such as Soft Switches, NAP (Network Access Point) addresses, security, service levels, etc.

This degree of granularity in measurements not only assures all bases are covered with regard to assessing service performance; it also provides the information an installer needs in order to know whether he can fix the problem or, if he needs to request help, whether the problem should be addressed by network ops or an IT technical unit.

To provide a full perspective on service performance measurements from all VoIP- specific interior outlets the House Check should tabulate: • MOS (Mean Opinion Scoring) – Figure 1: Handheld with House Check MOS is a measure of phone call Metrics quality and requires use of sophisticated algorithms in a software These measurements must be accessible to program that looks at several quality the installer via standard handheld devices parameters, such as jitter, latency and with user interfaces that clearly spell out the packet loss, to determine where the metrics with respect to each type of voice signal should be placed on the measurement. And measurements must be industry standard MOS scale. recorded and stored for immediate and • Stability – It’s essential that enough ongoing access from Customer Service measurements be made over time Representatives, Dispatchers, Engineers and across enough data points during NOC personnel. See Figure 1 for an installation to prevent over-reliance illustration of handheld User Interface. on what could turn out to be an anomalous performance reading taken The MTA, of course, is an essential on one data point at a single point in source of information about service time. performance. The House Check performed at • Inside wiring performance – The the MTA should measure and record: installer must be able to assess power • Levels and persistence of errors and signal-to-noise levels at every generated from the external network; outlet and take remedial steps to • Premises wiring noise, distinguishing adjust levels to compensate for the as to types such as ingress, device- impact of splitters and spurious noise generated noise and spurious noise from microreflections. from microreflections; • Transmit and receive power; In addition, the installer should have at his Beyond the initial install, the system must disposal a ready means of measuring and be able to generate real-time and historical recording levels at the tap and ground block. reports reflecting performance over time. These measures not only serve to identify With this level of comprehension the system whether there are local outside plant issues; should be able to isolate and prioritize they also create a clear benchmark for problems and provide quick and efficient gauging contributions of in-home wiring and recommendations to resolves the issues, both other elements to signal degradation at all at the time of install and subsequently VoIP-specific outlets. And the technician whenever problems arise. must be able to perform these measurements with respect to transmit and receive As part of the analytical process, the parameters across all VoIP-related DOCSIS House Check support system should be able channels. to go beyond generating raw performance metrics on a per-device, per-location basis to THE ENABLING SOFTWARE calculate a MOS score for the service at all INFRASTRUCTURE end points. Such MOS summaries measure network health by determining availability in Creating an easy-to-use, thorough House terms of Degraded Modem Hours (DMH) or Check process as described above requires a Severely Degraded Modem Hours (SDMH), well-conceived software infrastructure that and identify network issues that are causing fully exploits the data collecting power of QoS-sensitive applications such as voice to DOCSIS with complete compatibility with fail. multiple vendors and support for multiple DOCSIS and PacketCable devices, including The Full Scope of House Check Benefits CMTSs, call management servers, media gateway controllers, cable modems, DOCSIS There are many immediate and long-term set-tops and MTAs. To be useful beyond the benefits to be derived from implementing a installation stage, the software should House Check process like the one described interface with a complete VoIP lifecycle here. From the installation perspective, the management solution that optimizes on- process reduces the time spent by installers going fault and performance management. on each job and provides operations support staff a complete picture of post-install The system must be able to isolate and performance at the site versus what the resolve issues at the time of installation and performance looked like when the installer beyond by providing real-time and historical started running the analysis. data about premises devices, supporting interfaces and the CMTS. This includes the The process eliminates time-consuming ability to identify and resolve typical manual measurements and correlation of transport issues such as jitter, fragmentation, different types of performance parameters at latency and packet loss. The system must be the tap, ground block and premises outlets. able to draw together and interpret real-time Instant feedback minimizes time spent on and historical information for any given troubleshooting. And, if more work is premises device, providing detailed views to required, the process further cuts installer the NOC on location, configuration, settings work load by automatically creating and performance while delivering read-outs thorough maintenance work orders. of pertinent device information on the technician’s handheld at the customer site. Another enormous benefit is the ability to CONCLUSION quantitatively measure and report ‘success rate’ of installs by individual installer or Now that MSOs have made the business groups of installers/contractors. The most case for digital voice to where they are capable installers can be recognized and scaling service to millions of households, the given more ‘challenging’ installs and trouble bottom-line focus necessarily shifts to calls. And the less capable installers can finding ways to reduce costs and raise receive training in specific areas that are customer satisfaction. deficient. Gone are the days of wondering which staff are the most capable and which The opportunity to eliminate the primary staff repeatedly have install/trouble call cause for voluntary disconnects, namely difficulties. repeat trouble calls, while significantly lowering installation and ongoing operations Beyond the installation phase, the costs is one of the great benefits the cable information generated by the House Check industry derives from operating in the establishes a comprehensive data base which packet-based DOCSIS environment. Now, Customer Service Representatives, with the means at hand to support whole- Dispatchers, Engineers and NOC personnel home House Check functionality, operators can use to trouble shoot problems over time. can look forward to ongoing improvements For example, a shift in MTA readings that in the bottom-line performance of digital matches the readings taken at a secondary or voice. tertiary outlet during installation can indicate the device has been moved, allowing the References CSR to suggest it be put back in its original location to avoid a maintenance call. And, if i Undisclosed MSO Customer Data a truck roll is necessary, the comparative ii MSO-commissioned study readings can tell Customer Service Representatives, Dispatchers, Engineers and NOC personnel whether the problem is plant related or, if it’s on premises, what type of expertise and equipment will be required to perform the work.

The operational savings resulting from implementation of a House Check process are evident across several parameters. The process ensures a higher success rate for VoIP installs and virtually eliminates post- install trouble calls. Over time there are fewer complaints per customer, which increases customer satisfaction. And when there are problems, less time is spent trouble shooting and fixing them. Moreover, the accumulated performance record across service areas allows NOC personnel to discern patterns in the problem reports that can track problems geographically.

HYBRID IP-QAM VIDEO SOLUTIONS Christopher Poli, P.E., Richard Moore, William Weeks Motorola Connected Home Solutions

Abstract VOIP and internet applications continue to increase pressure on high speed data rates. A Hybrid IP-QAM System Architecture is one of the most effective Over the past couple years, at least architectures available today in expanding one major successful deployment of a plant capacity – allowing operators to offer national hybrid QAM IP architecture has an expanded and wide spectrum of high already occurred (and continues to expand). definition, international and niche Municipal entities are in the planning stages programming. The architecture is capable of or have already begun overbuilding as well – fully supporting Broadcast, Unicast or some are looking at a pure IPTV play and Narrowcast services (i.e., VOD) and Switched others at hybrid architectures that provide Digital Video; multiple video encoding both broadband IP and QAM delivery formats (e.g., MPEG-2, MPEG-4) and can capabilities. support a transition to an all IP STB solution. While a complete IP network build-out will There are options to increase the likely take years and millions of investment effective plant bandwidth capacity beyond dollars to complete; a hybrid system provides splitting nodes to the point of fiber exhaust. a bridging architecture without stranding Moving to 1 GHz for data services may also existing set-top assets. This paper provides provide some relief (video still up to 860 an overview of this architecture including a MHz). Going all digital – and recapturing comparison of the bandwidth delivery analog spectrum. Switched Digital Video is provided in hybrid architectures to that of an emerging technology as another means to other architectures. reclaim channel bandwidth – and combined with DOCSIS 3.0, could provide significant INTRODUCTION relief to forward bandwidth demand – both from a video services standpoint as well as The competitive landscape is changing data services. All of these activities certainly and becoming more challenging. Satellite help a great deal in the forward path, but they delivered TV has publicly announced plans to neither address the longer term need for launch 150 HD channels (source: DirecTV significantly more bandwidth nor the need for website) – nearly ten times the typical HD significant additional symmetrical bandwidth. offering of most cable MSO’s. It would take 75 QAM streams of cable plant capacity to This paper will cover an End to End carry that many HD services in MPEG-2 Network and System Architecture format. Competitors are deploying highly culminating in a PON based delivery to the competitive video services in many major home with hybrid IP and QAM capable set- markets with plans to offer richer content and tops. A hybrid IP QAM set-top has the more bandwidth than is possible in an HFC capability to receive services both on an IP plant. VOD capacity requirements continue interface and through the more traditional to grow – pushing nodes deeper into the cable QAM tuner. The set-top applications plant. Symmetrical applications such as determine the video source to be displayed or recorded. The IP interface can use either a traditional ethernet interface or a home In today’s MPEG-2 environment, network coax-based interface such as delivering 100 HD streams and their Multimedia over Cable Architecture (MoCA) equivalent SD streams will consume on or HomePNA. A home network based average 60 QAMs. That is nearly half of the architecture such as MoCA can provide 70- QAM channels available on an 860 MHz 100 Mb/s IP interface over an RF channel that HFC plant. Take away analog channels and runs above 860 MHz. A properly deployed, those dedicated to data and there is very little Hybrid architecture offers strategic bandwidth left to deliver compelling services advantages to the service provider. such as VOD that are capable of significantly reducing subscriber churn. While data The paper will also cover benefits of services may likely be moved above the 860 integrating Video on Demand or other MHz channel frequency to extend available singlecast services onto the IP data path, and bandwidth, it still requires pushing the nodes briefly describe the use of multicast on the IP further and further out into the plant to pipe to deliver carousel data, tunnel singlecast provide the needed bandwidth to deliver all data, or deliver IP video services. The paper the services desired. As nodes are split, fiber will finally cover how an IP capable exhaust may become a serious consideration architecture can enable seamlessness in in the long term. communications to a multitude of home and mobile devices – a two way IP pipe to stay How much bandwidth is required in connected to the world. building out a pure IP network that has the ability to deliver enough bandwidth to each BANDWIDTH NEEDS home to provide satisfactory service? Even How much bandwidth is enough? with the promised bandwidth advantages of With the proliferation of HD content and MPEG-4, high definition services will multiple HD TV sets in a single home, consume 8 Mbps/service. If we assume a bandwidth required to the home in either a peak requirement of simultaneously traditional HFC or Switched Digital delivery delivering two HD services, an SD service will drive the need for architectures that reach and 6 Mbps best effort data service, then the beyond simply pushing QAM nodes further peak BW per home is approximately 26 Mbps and further outward toward the edge. As the to a single home. Although some of the price of HDTV sets continues to fall, and the bandwidth is delivered to multiple homes in obvious quality difference between HDTV an IP Switched Digital Video deployment, and Standard Definition TV continues to rise that network will need to be managed – within the subscriber base, the need to deliver meaning that decisions to allow new video richer content – certainly beyond the typical sessions to traverse a particular path may be 10 to 20 HD services offered in an HFC plant dependent on current bandwidth consumption - will rise significantly in the near future. Add on that path. Depending on specific market to that the ever increasing demand for data assumptions, the 26 Mbps/home number may services bandwidth and it may become be adjusted upward or downward. It certainly apparent that subscriber demand will exceed would be no less than 20 Mbps/home and an today’s HFC QAM channel plant capacities. argument might be made for as much as 40 The cable provider who is prepared to deliver Mbps/home or more. While this number may the bandwidth to enable these services and seem absurdly high today, how long ago did more will have a significant competitive 750 Kbps DSL, 1.5 Mbps DSL or 6 Mbps advantage over the operator who cannot. Cable Modem rate seem absurdly high when phone line data connections were only the description above points out what is capable of delivering 56kbps? Data already available on the QAM side of the customers moved very quickly from slower hybrid architecture – including the application data service providers to high speed cable management layer. This HFC foundation can modems. The future may hold another be expanded by adding in the PON migration to competing services that offer architecture with only minor changes to the more or business-necessary erosion to service existing video infrastructure. margins to keep the current customer base. HYBRID QAM-IP ARCHITECTURE Today’s HFC services compete for available bandwidth. In a typical HFC A QAM-IP architecture enables architecture, QAM channels are allocated existing, previously integrated QAM based across multiple service types. As bandwidth applications and augments that plant capacity demands increase, HFC architecture may with an IP infrastructure that can grow as move toward Switched Digital Video where a demand for bandwidth grows. Deployed number of QAM channels are allocated to applications can easily migrate to utilize the switched broadcast (essentially multicast), IP back channel as opposed to traditional thin- singlecast (i.e., VOD) or data services. pipe RF return. Depending on the allocation and number of subscribers served per node, these nodes will Hybrid architectures provide push deeper and deeper into the network significantly more forward broadcast and requiring more and more fiber to provide the narrowcast capacity than non-hybrid more effective bandwidth per subscriber to architectures. At the same time, density of remain competitive. homes per GigE served can be increased since the majority of forward video broadcast load can be delivered on the QAM portion of the plant. A hybrid QAM IP architecture using a typical GPON delivery can offer a full 860 MHz QAM plant (or 1 GHz if data rides above 860 MHz). In addition, the OLT acts as a multicasting engine and switch that allows 32 users to share up 2.4 Gbps IP downstream and 1.2 Gbps IP upstream.

A typical Hybrid IP-QAM PON Architecture utilizes a dedicated wavelength for the traditional 860 MHz cable plant services and separate dedicated wavelengths Figure 1: Typical HFC Services for the IP connection upstream and downstream. These can be carried on the This paper is not aimed at converting same fiber or on different fibers to a node from a typical HFC architecture to PON accumulation point or to the side of a home. architecture as conversion may have an Figure 2 illustrates a system deployment to insurmountably high network infrastructure, the side of a home. headend and customer premise cost and does not adequately leverage the enormous investment in the current QAM plant. Rather, The Headend or Supernode has typical configuration services). Services are cable headend functions including derivation received, processed (may include ad of broadcast video services, VOD (may also insertion), encrypted (as required), RF be delivered directly on IP path), IPPV combined and transmitted optically on 1550 services, and out of band messaging (may nm wavelength. Data services including include CAS, EPG, and set-top related

Voice OLT & 1550 Services PON Overlay 1310nm A single PON Edge supports up to 32 Router 1490nm subscriber IP Cloud or OLT WDM premises Edge Mux

Switch Passive 1550nm Split Optical IP Data TV Services Traditional Set-top HFC Plant TV Optical Transmitter Set-top 50MHz to 860MHz Headend or RF coax Settop RF Combiner VOD System Super Node Reportback ONT (signalling) (IPPV) Ethernet Broadcast Video Cat5 OOB Router Encrypt, OOB Mux Multiplex, QAM Modulate Conditional IP Telephone Access Receive/Encode Video Subscriber Settop Settop Applications VOD System Premise Polling PC (IPPV) Data (EPG (video streams) Carousel)

Figure 2: Functional Hybrid PON Network – IP Return Channel

VOIP and optionally video services such as Q-IP set-top is eliminated/not required. All VOD can be delivered on a separate IP path, connectivity can be provided through an IP transmitted optically on 1490 nm wavelength. network connected through an upstream edge 1550 and 1490 wavelengths are wave division switch/router to the OLT. multiplexed onto the PON at the output of the OLT. The return path from the ONT to the In figure 3, the OOB multiplex, set- OLT is IP on 1310 nm and no RF top application data and VOD system are demodulators are required to receive upstream moved completely over to the IP data path. In transmission from the set-top. In addition, addition, new services such as streaming application communications downstream may video and even switched video can be also traverse the IP path and an OOB provided across the IP path. Lastly, all return communication channel downstream (such as channel functions are also moved to the IP utilized with some RF return networks) is not path. Carousel data is easily handled with the required. Specialized headend equipment to multicast capability of GPON. Specific enable communication from application requirements for IP multicast data carousel server in headend to application client on the would likely be application dependent. Application signaling (such as required for VOD) and application service delivery is The subscriber premise as shown in typically singlecast to the client. By both figures 2 and 3 assume a home IP providing for both QAM and IP paths, network over coax such as Multimedia over bandwidth utilization – particularly on the Cable Architecture (MoCA) or HomePNA forward path – can be optimized, and which provides return path channel and flexibility in the manner in which services are delivered is maintained. Voice OLT & 1550 Services PON Overlay Data 1310nm A single PON Services Edge supports up to 32 Router 1490nm subscriber IP Cloud or OLT WDM premises Edge 1550nm Mux Switch Passive Split Optical TV

s

g m n i Set-top ea r t nal s g i o S e TV d i V Optical Transmitter Set-top Switched 50MHz to Video 860MHz RF coax RF Combiner Settop ONT VOD Streaming Ethernet Reportback System Video Headend or Cat5 Router (IPPV) Broadcast Video Super Node Set-top OOB Mux Encrypt, Applications Multiplex, QAM IP Telephone Data (EPG) Modulate Subscriber Conditional Premise Settop Access Receive/Encode Polling PC (IPPV) Video

Figure 3: Functional Hybrid PON Network – IP Services

enables advanced features like whole home Specific deployment requirements network or multi-room DVR. MR-DVR will may vary based on existing operator allow video to be selected and streamed from infrastructure and longer term desires. a DVR set- top to a non-DVR set-top on the Considerations may include existing network same home network. Set-tops may be and equipment deployments, fiber provisioned behind a home router or directly availability, provisioning requirements, and from the network / system. Ideally, set-tops the like. Benefits of integrating Video on sit behind a home router and are provisioned Demand or other singlecast services onto the locally on the home LAN. DHCP and DNS IP data path, and the use of multicast on the functions are used for IP provisioning and IP path to deliver data or video services application client-server connection. provides more bandwidth on the QAM side of Standard IP protocols govern both the hybrid architecture which will allow high provisioning and communications. bandwidth HD broadcast services to be delivered without consuming all important In a seamless mobility solution, bandwidth for future applications on the IP operators collectively deliver a continuous side. Lastly, an IP home network can more experience of content. Using a network readily enable seamlessness in management protocol such as IMS, switching communications using an IP network to a Cable, Wireline and Wireless sessions occurs multitude of home and mobile devices – a two at the session layer. The transport layer – way IP pipe to stay connected to the world. meaning any transport layer, including a network managed IP or QAM broadcast PROS AND CONS service – provides transport services only. Applications become transport indifferent. Reasons for deploying PON based Subscriber access points may be Cable, architecture include substantially enhancing Data/IP (FTTP), Cellular, WiFi, or WiMax, bandwidth to the home easing future but are driven by the subscriber and bandwidth constraints, using it in targeted associated applications. In an ethernet areas where competitors are focusing, and centric world, a bandwidth rich IP lower maintenance costs than an active and infrastructure is absolutely necessary to stay largely metallic network. FTTP certainly competitive and the additional QAM transport must be considered for greenfield layer provides downstream capacity that deployments. Challenges include costly offloads the IP infrastructure substantially. rebuilds that may not be popular in the near term to stockholders and potentially stranding Bottom line - Hybrid architectures some deployed investment assets as new allow the operator to be ready to fully enable technology is deployed. the flexible Connected Home.

VISION FOR THE FUTURE

An IP enabled cable plant allows for more diverse service offerings as well as provides a framework for the future. Transcoding to formats such as MPEG-4 in the headend will allow for more services in less bandwidth. It is expected that MPEG-4 content would be delivered over IP simply because it is assumed the receiving consumer device will have an IP interface. Transcoding content in the HE may also enable delivery over other mediums such as wireless/WiFi/cellular/DVB-H networks. In such a seamless network, IP video delivery requires the creation of SPTS, individually transcoded and encrypted and either stored on a headend server or streamed directly to the mobile user.

INTRODUCING LcWDM™ – THE NEXT WDM TECHNOLOGY FOR THE CABLE INDUSTRY Oleh J. Sniezko, Sudhesh Mysore, Charles Barker Aurora Networks, Inc.

Abstract INTRODUCTION

This paper presents a WDM technology Quest for Bandwidth Never Stops for downstream HFC communication. The technology, trademarked as LcWDM™, is a The demand for bandwidth among the dense wavelength division multiplexing customers of the telecommunications (DWDM) technology based on an extension network operators increases continuously. of the ITU-T Recommendation G694.11 to On the other hand, the telecommunications the optical O-Band (1260 to 1360 nm). network operators today are looking at adding services and programming to increase An order of magnitude decrease in revenue potential and to match the demand wavelength spacing (as compared to for HDTV channels. This trend includes CWDM) allows all LcWDM™ wavelengths addition of bandwidth capacity on fiber to to be within ±20 nm window about the serve residential and commercial customers. nominal zero dispersion wavelength (λ0) of HFC broadband network operators are also standard SMF-28 type fiber (ITU-T G.652). facing competition from satellite operators in This greatly reduces chirp-induced CSO due video services and from Telcos in all to fiber dispersion and allows for longer services. Two of the most leading fiber spans. challengers from this side are Verizon (FiOS) and AT&T (U-Verse). This paper describes the various fiber phenomena – both linear and nonlinear – The continuous demand for increased that limit the fiber distance and the number bandwidth capacity per user is forcing of wavelengths that can be supported by the segmentation of the node areas into smaller two technologies (CWDM and LcWDM™). service groups. In parallel to this trend, the Descriptions of the detailed testing of fiber expansion of urban areas force HFC new nonlinearities and how they affect system builds with many nodes deployed every year performance as well as architectural to serve areas that were not served applications of the LcWDM™ system is previously. Both of these trends (demand for also presented. higher bandwidth per user and demographical expansion of urban areas) The LcWDM™ system has been lead to increased demand for fiber capacity carefully engineered to ensure that the fiber through new construction or more efficient dispersion and nonlinearities of a typical fiber capacity utilization. The first choice, (SMF-28) fiber, prevalent in HFC optical new fiber construction, may prove costly in node links, do not degrade the link areas already equipped with fiber (at least to performance below the acceptable levels. the existing serving area boundaries) but which lack dark fiber. Even at modest construction cost per mile and average population density, the cost per household (served or not served) will amount to $150 downstream HFC communication are and more. For downtown areas, this cost can CWDM and LcWDM™. These technologies be much higher, even prohibitively so. try to close the gap between the two extremes described above. The second choice is preferred as it is an order of magnitude lower in cost than the The coarse wavelength division cost of new fiber construction. There are multiplexing (CWDM) technology was several technologies that improve the introduced to increase efficiency of fiber efficiency of fiber utilization. These utilization in shorter fiber runs and where the technologies can help to free up fiber on fully lower number of wavelengths per fiber was occupied optical cable routes for sufficient. CWDM wavelengths are segmentation, serving area expansion and specified in an ITU-T Recommendation even new service addition (SMB packages). G.694.22. This document specifies up to 18 wavelengths between 1271 nm and 1611 nm The most effective and mature with 20 nm wavelength spacing. In most technology that has been deployed by many applications on older fiber with higher water- operators is C-Band DWDM (1530-1565 nm) peak loss, only 15 wavelengths can be used. and its distributed DWDM version. This technology allows for practically unlimited This technology has been successfully narrowcast bandwidth (upper octave in any used in the upstream path, both analog (40km system, e.g., upper 500 MHz of narrowcast links) and digital (where it is not distance signal in 1002 MHz systems) on 40-plus limited to 40 km) and for digital downstream, wavelengths over a single fiber, providing Ethernet links (data, T1 over IP supplemented by an additional broadcast and VoIP) for SMB services. wavelength on the same fiber. In addition to its highly efficient fiber utilization, this Downstream CWDM versus LcWDM™ technology allows for practically unlimited Distance/Capacity Limits distance (100 plus km up to the DOCSIS latency limits). In downstream communication, where the optical transmitters are used for DWDM technology, aided by TDM in transporting SCM (sub-carrier multiplexed) digital reverse links, has been used for analog video and QAM signals, two major increasing the efficiency of fiber utilization challenges significantly limit the use of for upstream communication in long-distance CWDM technology3. One of them is the or on extremely fiber-starved routes. high level of dispersion (see Figure 1) in SMF-28 fiber at CWDM wavelengths other At the other extreme of the fiber than 1311 nm. This fiber type has been utilization efficiency spectrum are single- dominant in access plant deployment in HFC wavelength systems: either 1310 nm directly networks. The high dispersion, resulting modulated DFB laser links in downstream from the large 20 nm CWDM wavelength and upstream transport, which have limited spacing, combined with the chirp of directly distance of 40 km if no passive loss is modulated laser transmitters, results in high present, or 1550 nm eternally modulated levels of CSO. The other technical hurdle is laser links in downstream communication. the high level of crosstalk between CWDM wavelengths on the same fiber due to SRS The most recent WDM technologies that (see Figure 2) and XPM phenomena. are being introduced by vendors for 25.00 the allowed fiber distance (loss budget) in 20.00 CWDM systems with more than two 15.00 wavelengths. 10.00 1. 2 0

5.00

1. 0 0 0.00 1270 1290 1310 1330 1350 1370 1390 1410 1430 1450 1470 1490 1510 1530 1550 1570 1590 1610 Dispersion [ps/(nm-km)] Dispersion -5.00 0.80

-10.00 Wavelength [nm] 0.60

Figure 1: Dispersion Characteristics of 0.40 SMF-28 Fiber Raman Gain [x10exp(-13)m/W] 0.20

High dispersion limits the fiber span 0.00 0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0 45.0 lengths to 15 km even for the two CWDM Frequency Separation [THz] wavelengths closest to the nominal zero- Pump@1530nm Pump@1310nm Pump@1000nm dispersion wavelength. This limit can be a) Raman Gain vs. Wavelength Separation increased only with sophisticated and 1. 2 0 complex dispersion compensation and/or mitigation technologies, which increase the 1. 0 0 cost of the transmitter. Use of very low chirp 0.80 lasers, dispersion pre-distortion techniques (in distance intervals) and passive dispersion 0.60 compensation techniques are a few examples. 0.40

However, these are even more complex than Raman Gain [x10exp(-13)m/W] techniques used in the DWDM systems 0.20 described above as they have to be effective 0.00 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 with analog channel load across the entire Wavelength Separation [nm] forward RF spectrum, which spans several Pump@1530nm Pump@1310nm Pump@1000nm frequency octaves. Moreover, some of the b) Raman Gain vs. Frequency Separation remedies increase sensitivity to other fiber Based on G.P. Agrawal, Nonlinear Fiber optics, 3rd ed. induced impairments that require additional (Wiley, New York, 2001), adapted from R.H. Stolen and E.P. Ippen, Appl. Phys. Lett., 22, 276 (1973). techniques to neutralize (e.g., low chirp lasers are subject to lower SBS threshold and Figure 2: Raman Gain for Three Different higher IIN unless SBS suppression and IIN Pump Wavelengths neutralization techniques are deployed). An alternative technology, trademarked SRS-induced crosstalk limits the total as LcWDM™, is based on extension of the power allowed into the fiber for wavelength ITU T.G694.1 standard to the optical O-Band separations larger than 20 nm. Note that in a (1260-1360nm). All wavelengths used in three-wavelength CWDM system, the this system are within a ±20 nm window separation between the two extreme about the nominal zero dispersion wavelengths is 40 nm. Since SRS crosstalk wavelength (λ0) of standard SMF-28 type increases in severity for larger wavelength fiber (ITU-T G.652). Each set of separations (up to 13 THz frequency wavelengths is engineered for a total separation – equivalent to about 80 nm difference between the extreme wavelengths wavelength separation at 1310 nm and 110 not to exceed 15 nm. This significantly nm wavelength separation at 1550 nm), this lowers the rate of dispersion accumulation affects CWDM systems and further limits with distance and results in lower dispersion- The phenomena listed above are well induced CSO in longer fiber spans. described in the literature. This paper will Moreover, by placing all wavelengths closer focus on the impact they have on the together, SRS crosstalk is minimized even at performance of different optical link low RF frequencies. technologies and then concentrate on the LcWDM™ technology based system IMPAIREMNTS IN OPTICAL LINKS performance in presence of those CAUSED BY LINEAR AND NONLINEAR phenomena. FIBER PHENOMENA The dominant impairments caused by Nonlinear and Linear Fiber Phenomena fiber in multiwavelength systems are crosstalk and second order RF distortions Downstream HFC optical transport links (CSO). The challenge in designing the involving analog video and digital video multiwavelength system and its components QAM SCM (subcarrier multiplexed) RF (active and passive) is to minimize these two channels carry signals that are extremely impairments while maintaining acceptable sensitive to fiber effects that give rise to CNR, CTB and BER/MER performance of noise (CNR degradation) and nonlinear the transported signals. This is achieved by signal distortions − primarily second order, balancing the effects listed above. which results in CSO degradation. System degradation to varying degrees is observed Fiber Phenomena and Multiwavelength due to the following linear and nonlinear Optical Transport Systems effects: 1. Fiber Chromatic Dispersion (CD) As mentioned above, the three major multiwavelength systems available (some of 2. Interferometric Intensity Noise (IIN) them widely deployed) for downstream links to optical nodes are:

3. Fiber Nonlinearities 1. 1550 nm broadcast with externally a. Inelastic Scattering with Phonons modulated lasers, together with 1550 i. SBS (Stimulated Brillouin Scattering) DWDM narrowcast overlay on directly ii. SRS (Stimulated Raman Scattering) modulated lasers

b. Nonlinear Refractive Index 2. CWDM directly modulated lasers i. Cross-Phase Modulation (XPM) ii. Self-phase Modulation (SPM) iii. Four Wave Mixing (4WM) 3. LcWDM™ directly modulated lasers iv. Optical Kerr Effect and Polarization Dependent Loss (OKE/PDL) Each of these systems is affected differently by fiber phenomena. Table 1 4. Higher-order interactions of the above compares these impacts and lists some phenomena remedies implemented in these systems to limit or neutralize the contribution of the fiber phenomena to system performance degradation. Table 1: Impact of Linear and Nonlinear Fiber Phenomena on Different Optical Transport Systems to Optical Nodes Fiber Its Impact on Phenomenon 1550 BC/1550 DWDM NC Overlay CWDM LcWDM™ 1550 nm BC 1550 nm NC (40 wavelengths) Chromatic No real impact Significant impact Significant impact No significant impact Dispersion unless for very neutralized by (no simple remedies, if laser chirp is long distance (due frequency allocation low chirp lasers may optimized for both to SBS and/or dispersion require SBS chromatic dispersion suppression predistortion (cost suppression circuitry levels and fiber IIN system) impact). Other and IIN suppression methods (e.g., dithering) selection of low chirp lasers) may require additional remedies. IIN No impact (IIN is The same order of The same order of The same order of below the lower magnitude as for magnitude as for magnitude as for forward directly modulated directly modulated directly modulated bandwidth 1310 nm DFB lasers 1310 nm DFB lasers 1310 nm DFB lasers frequency) unless low chirp lasers are selected to lower the impact of chromatic dispersion SBS Neutralized to the No impact for No impact for No impact for SBS threshold by standard chirp lasers standard chirp lasers standard chirp lasers SBS suppression in the range of optical in the range of in the range of optical methods fiber launch powers optical fiber launch fiber launch powers powers SRS NA (other No significant impact Significant impact No significant impact wavelengths have for QAM channels. for 3 and higher on QAM channels different RF number of and analog channels frequencies) wavelengths per as long as analog fiber channels are the same XPM NA (other Not a dominant source Not a dominant Contributes to wavelengths have of crosstalk source of crosstalk crosstalk for lower different RF separation frequencies) wavelengths 4WM No impact No impact Limited impact Measurable impact (under very unlikely neutralized by the scenario) system design

MECHANICS OF DEGRADATION OF characteristic of SMF-28 fiber is shown in OPTICAL LINK PERFORMANCE Figure 1.

Chromatic Dispersion and Related Chirp Variations in group velocity with Penalties wavelength combined with periodic wavelength shifts due to laser chirp in a Chromatic dispersion refers to the directly modulated transmitter results in CSO variation in group velocity with wavelength. distortion of analog signals4. This This characteristic is well-described by the dispersion/chirp-induced CSO increases very Sellmeier equation, and a typical dispersion rapidly with dispersion, even for a 20km system, as shown in Figure 3. Several plots increases as the frequency difference are shown for laser chirp (at 100% OMI between the two interacting signals modulation) ranging from 0.5 GHz to 10 increases, and reaches a peak at around 13 GHz. THz (see Figure 2a). Although the gain peak

-30.0 occurs at a fixed frequency difference of 13 -40.0 THz when plotted against frequency shift, -50.0 the position of the peaks vary for different -60.0 pump wavelengths (i.e., the shorter of the CSO [dBc] -70.0 two interacting wavelengths) when plotted -80.0

-90.0 against wavelength shift as in Figure 2b. The 1270 1275 1280 1285 1290 1295 1300 1305 1310 1315 1320 1325 1330 1335 1340 1345 1350 Raman gain increases for wavelength Wavelength [nm] separation up to 80 nm for a pump near 1310 10 GHz 5 GHz 3 GHz 1 GHz 0.5 GHz Figure 3: Chromatic Dispersion nm, and up to 110 nm for a pump near 1550 Accumulation with Wavelength (20 km of nm. Furthermore, the amplitude of the SMF-28 Fiber) Raman gain decreases according to well- known scaling rules as the pump wavelength Note that even with CSO specifications moves towards longer wavelengths as shown relaxed to −60 dBc, CWDM systems are by the three curves in Figure 2b. limited to three wavelengths (1291nm, 1311nm and 1331nm) due to SRS-induced crosstalk can be extremely dispersion/chirp-induced CSO for a typical high at low RF frequencies as shown in 100% OMI laser chirp of 5 GHz for 8 dBm Figure 4 for 55 MHz. For typical system lasers. This is under the assumption that the parameters (+10 dBm/ch launch power, 1310 nm pump, 25 km fiber distance) the SRS- average zero dispersion wavelength λ0 in the link is equal to the nominal zero dispersion induced crosstalk can exceed −40 dBc for wavelength. The CSO will degrade further wavelength separations larger than 30 nm. for one of the CWDM wavelengths if the λ0 Raman Crosstalk (55 MHz, 1310nm Pump, in the link is offset from the nominal value. +10 dBm, 25.3km) Moreover, this CSO will cascade with optical -25.0 link CSO generated by other mechanisms -30.0 (including laser transmitter CSO). -35.0 -40.0 Crosstalk Due to Stimulated Raman -45.0 Scattering -50.0 -55.0

(dBc) Crosstalk Raman -60.0 SRS refers to optical gain experienced by 0 20 40 60 80 100 120 140 160 one wavelength signal at the expense of Wavelength Separation (nm) another shorter-wavelength signal. Although the transfer of optical power is from the Figure 4: SRS-Induced Crosstalk short-wavelength signal to the longer- wavelength signal, this transfer is modulated by the product of the optical power of both wavelengths, leading to leakage (crosstalk) of the RF signals in both directions at almost the same level. The coefficient of the product term is denoted by g, the Raman gain coefficient. The Raman gain coefficient Cross-Phase Modulation (XPM) 1. All analog signals on multiple wavelengths on the same fiber must be At high RF frequencies (around 500 the same (the same content at the same MHz) and small wavelength separation, the frequencies). This limitation allows for crosstalk does not vanish as predicted by lowering the requirements for crosstalk Figure 4 due to the effect of another fiber levels. nonlinearity, namely cross-phase modulation (XPM). This nonlinearity is caused by the 2. Even for the same signals, multipath nonlinear refractive index – modulation of effect must be accounted for (from Figure the fiber refractive index – which causes 5, for −35 dBc of crosstalk, only 100 to modulation of one wavelength to induce 250 ns differential delay is allowed for phase changes in the other wavelengths. This signals traveling on different wavelengths induced phase modulation in conjunction on the same fiber – this translates to 80 to with fiber dispersion results in crosstalk from 200 feet of a typical RF cable in the one wavelength to another. headend combining network).

Fiber dispersion is therefore a necessary Four Wave Mixing (4WM) ingredient for XPM-induced crosstalk; XPM- induced crosstalk increases with increasing Four-wave-mixing is another result of the fiber dispersion and decreasing wavelength fiber refractive index nonlinearity that results separations. Like other refractive index in three optical signals at distinct frequencies nonlinearities, XPM-induced crosstalk fi, fj, and fk interacting as they propagate depends strongly on the group velocity along a fiber to give rise to a fourth optical mismatch (or “walkoff”) between the signal at frequency fijk = fi+fj−fk – similar to affected wavelengths, which is approximated a third-order “beat product” in RF systems. by the product of the fiber dispersion times The mixing product at frequency fijk is said to the wavelength separation. be non-degenerate if i≠j≠k (i.e., the mixing product is generated by three distinct The actual crosstalk observed in a signals). It is possible for 4WM to occur multiwavelength system depends on the with only two optical signals present (at relative magnitude and phase of the SRS- frequencies fi and fj), giving rise to a so- 5 induced and XPM-induced crosstalk . While called “partially-degenerate 4WM” optical crosstalk from other wavelengths (PD4WM) products at frequencies fiij and fjji. appear at the same location (under each analog carrier) as CTB distortion (which The mixing efficiency of the 4WM arise from other RF carriers in the same process is maximized when the phase- optical wavelength), the CTB can be mismatch parameter is zero. The relative differentiated from the crosstalk as it is 4WM power Pijk/Pout generated by three composed of a large number of discrete RF polarization-aligned signals with the same beat products that are slightly offset from fiber launch power Pin and fiber output each other. power Pout can be easily calculated from a simple equation6. The same slightly At this high level of crosstalk (see SRS modified equation can be used for the case of crosstalk in Figure 4 and XPM crosstalk PD4WM mixing products. addition), two major network design limitations must be observed: The equations for the mixing products are valid only so long as they are small enough that the “pump power” in the original signals at power levels of +10 dBm/ch if the fiber is not depleted. Even so, they predict that dispersion is low since the corresponding very large mixing products can be generated phase-mismatch parameter is then also small.

100 to 250 ns (24 to 60 m of coax) 0

-5

-10

-15

-20

-25

-30

Delayed Signl [dB] -35

-40

-45 Required Ratio of between Primary and -50 10 100 1000 10000 100000 Delay Difference [ns]

Old Mertz Curve Luminance Echo Chrominance Echo

Figure 5: Echo Tolerance Curves for Color TV Programming Displayed on Large Screen (Based on Rogers Published Test Results – Gary Chan and Nick Hamilton-Pierce and in Line with Dan Pike’s7 Published Test Results)

Figure 6 shows the calculated power levels of 4WM and PD4WM mixing -10 products as a function of the phase-mismatch -20 parameter for +10 dBm/ch launch power and -30 20 km fiber distance. -40 -50

Note that relative mixing products as RF Egress (dB) -60 high as −20 dBc can be generated if the -70 system is not designed properly. Figures 7a -80 and 7b show the input and output spectra, -90 respectively, of an actual 5-wavelength 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 system, demonstrating that such high levels Phase Difference (radians/km) of 4WM and PD4WM can indeed be 4WM (Pijk/Pout) PD4WM (Piik/Pout) Pump Depletion Limit obtained in not optimized systems. Figure 6: Dependence of 4WM and PD4WM products on phase-mismatch parameter. term is composed of 79 analog channels, egress of power out of a wavelength generates a multitude of second and third order intermodulation distortion products and degrades the CSO (and to a lesser extent, the CTB) performance of the wavelength.

LcWDM™ SYSTEM TEST RESULTS

Test Setup

LcWDM™ systems with two to eight wavelengths were tested in a setup similar to a) Input Signals that presented in Figure 8. Each transmitter is loaded with 79 channels of CW carriers from 54 to 552 MHz and 75 channels of 256- QAM signals from 552 1002 MHz. Matrix 1 is used to drive the wavelength under test while Matrix 2 drives the other transmitters with other wavelengths (all wavelengths are tested for performance). This is necessary to measure crosstalk from the other optical wavelengths due to SRS, 4WM and XPM. If it is desired to test only the CTB for the particular wavelength, then the same signal source can be connected to all transmitters.

A polarization controller is attached to b) Output Signal Spectrum each transmitter and the worst-case settings Figure 7: Spectra of 5-Wavelength System are found prior to each measurement. with High Levels of 4WM and PD4WM Testing was performed on several dozen Products spools of fiber covering a wide range of fiber

parameters. With N=5 wavelengths, there are a total of N2(N-1)/2 = 50 mixing products Initial testing started with just two generated. Even though it is possible to wavelengths in order to test the basic theory. design the system so that none of the mixing With just two wavelengths present, it is quite products fall on any transmission wavelength easy to model the performance of each and hence there is no crosstalk caused by wavelength. This allows calculating the ingress of 4WM power into any wavelength, strength of each mixing products, in which there will still be egress of power out of each the particular wavelength is involved and wavelength. hence the CSO and CTB degradation

contribution of each of them. The baseline In fact, each wavelength is supplying performance of each single wavelength power into a total of (N-1)∗(3N-2)/2 = 26 without fiber is used as a reference. mixing products. Since this egress power is proportional to PiPjPk, where each of the P Opt. filter Optical filter DD CC BB AA DD CC BB AA

AT3312G-DD

P Polarizer

Matrix 2 50-550 CW AT3312G-CC for AR3002G loading the P other Txs with different analog content AT3312G-BB Analog video P

Matrix 1 50-550 CW AT3312G-AA Spectrum for Filter Bank CNR, CTB, P Analyzer CSO, Xmod testing

De-Correlated Analog video QAM Signals

BER Receiver

256-QAM generator 552-1002 QAM

TC Receiver

Video Modulator and Up-Converter

Figure 8: Test Setup for Multi-Wavelength LcWDM™ System

Figure 9 shows a 3-D surface that describes the strength of the PD4WM products for a system where one channel is fixed at λ2=1308 nm. The x-axis is the zero- dispersion-wavelength λ0 and the y-axis is the wavelength λ1 of the second channel. The fiber distance has been assumed to be 25 km and the power level to be +10 dBm/ch.

Note that as the phase-mismatch approaches zero, PD4WM product power rises sharply. Also, when wavelength separation is small, the PD4WM power remains high over a much larger range. Figure 9: Modeling PD4WM Power Figure 10 shows the results of testing made as λ1 is varied from 1307 nm to 1314 with a fixed wavelength at λ2=1311.5 nm. nm. Again, there is very good correlation The x-axis is the wavelength λ1 of the second between theory and measurements. Note channel, which is varied from 1307 nm to also that the PD4WM power level remains 1314 nm. Good correlation is seen between high over a much wider range of λ1 values. the observed PD4WM power and the calculated values. Tests on 3-, 4-, 5- and 6-channel systems also showed very good agreement with Figure 11 shows the same system but theoretical models. The tests will be also over a different fiber. Measurements are expanded to 8 wavelengths.

-5 -5

-15 -15

-25 -25

-35 -35

-45 -45

-55 -55 Beat Product Amplitude [dBc] Amplitude Product Beat Distortion Levels [dBc] Levels Distortion -65 -65

-75 -75 1307 1308 1309 1310 1311 1312 1313 1314 1315 Signal Wavelength [nm]

Calculated Beat Measured Beat Measured CSO Measured CTB

Figure 10: PD4WM Power versus Wavelength Separation on Fiber with Distant λ0

-5 -5

-15 -15

-25 -25

-35 -35

-45 -45

-55 -55

-65 -65 Distortion Levels [dBc] Beat Product Amplitude [dBc] Amplitude Product Beat -75 -75 1307 1308 1309 1310 1311 1312 1313 1314 1315 Signal Wavelength [nm]

Calculated Beat Measured Beat Measured CSO Measured CTB

Figure 11: PD4WM Power versus Wavelength Separation on Fiber with Close λ0

LcWDM™ SYSTEM DESIGN 2. XPM CHALLENGES 3. 4WM ingress

The LcWDM™ design took into account The optimization of the first two is the theoretical modeling results and difficult as the requirements to minimize extensive testing results of the nonlinear and SRS crosstalk most of the time conflict with linear fiber phenomena and their effect on the requirements to minimize XPM crosstalk. analog and digital QAM signal performance Fortunately, SRS crosstalk is highest at lower transmitted on multiple wavelengths. Two RF frequencies and XPM crosstalk is highest major parameters that posed the design at highest RF frequencies. Moreover, challenges were crosstalk and second order LcWDM™ system has a significant (CSO) distortions. The LcWDM™ system advantage as the separation between was optimized to meet the requirements for wavelengths are relatively small (in these two parameters while securing comparison to CWDM system) and hence acceptable performance for CNR, CTB and SRS crosstalk is low while XPM crosstalk MER/BER. Achieving this goal required affects mostly QAM channels. As long as balancing the fiber effects described the two assumptions: previously8. 1. All analog signals on multiple wavelengths on the same fiber are the Crosstalk same, and 2. Multipath effect is accounted for; Three major fiber phenomena cause (as explained above) are met, these two inter-wavelength crosstalk: effects do not limit the application of the 1. SRS system. Moreover, the third effect is completely eliminated by LcWDM™ system degradation is negligent (see Figures 10 and design. Additional contributors to crosstalk 11). (e.g., OKE-PDL interaction) are being controlled by system component selection APPLICATION OF LcWDM™ SYSTEM and specification. FOR NODE SEGMENTATION AND SERVICE AREA GROWTH Second Order Distortions (CSO) Figures 12 illustrates the use of Two major phenomena contribute to LcWDM™ technology for downstream degradation of the CSO in analog transport in conjunction with CWDM multiwavelength links: technology and digital technology for 1. Linear effect of chromatic dispersion upstream transport to segment an initial node resulting in chirp-related penalty, and into 5 forward and reverse segments using 2. 4WM egress. only two fibers. Additional node is added to split a node leg that served a high number of The first effect is minimized in households. Sixth node is added (in future) LcWDM™ system by the proximity of the for the new growth area. The upstream fiber wavelengths to the zero-dispersion- is not shown here but at the discretion of the wavelength λ0 of SMF-28 fiber that is operator, the same fiber that is used for predominant in optical links to the nodes. downstream transport can be used for Moreover, the laser chirp is managed to upstream transport (see Figure 13). optimize both CSO and IIN contribution. Alternatively, the second fiber can be used The second effect is minimized by the system for upstream transport (this will yield longer design. The 4WM egress in extreme cases reach as combining and de-combining filter causes CTB degradation but if the CSO loss is eliminated from the downstream degradation is under control, CTB path).

Initial Area of Optical Node

H H 1310 TX L L

SMF-28 Fiber H H Technology L L AT3310 (or equivalent) DT4030N

a) Initial Node Service Area Node Added to

H H Segment High LcWDM™ L L Forward Transport HHP Leg H H H H System L L L L Area H H AA Segmented L L BB CC into 5 DD DT4030N DT4230N Forward EE DT4230N FF Segments

H H L L Node Added to

H H Serve New L L Subdivision DT4030N b) Node Service Area Segmented into Fiver Areas with 6th Node Added to Serve New Subdivision

Figure 12: LcWDM™ System Applied for Area Segmentation and Service Area Expansion

LcWDM™ H H L L Area Segmented Forward Transport H H into 8 Forward System L L Wavelength 1 and 8 Reverse

Wavelength 2 -0.5 -0.5 Segments Wavelength 3 -1.0 -1.0 Wavelength 4

Wavelength 5 DT4230N Wavelength 6 DT4230N Wavelength 7 OP91S4S-EQ Wavelength 8

H H GR L L

BP3104 DR3021 H H DR3021 -6.6 L L DR3021

DR3021

DT4230N DT4230N

Figure 13: LcWDM™ Downstream System in Combination with CWDM/Digital Upstream System to Serve 8 Forward and 8 Reverse Segments on Single Fiber

CONCLUSIONS technology for narrowcast overlay. They are much more immune to both SRS-induced LcWDM™ systems close the crosstalk and dispersion-induced CSO than performance and cost gap between single CWDM systems. This allows for extending wavelength optical links based on 1310 nm their reach and increasing the number of directly modulated DFB laser technology and wavelengths per fiber. The fiber length DWDM 1550 nm systems based on dispersion limits are doubled. The externally modulated technology for LcWDM™ systems allow for cost-efficient broadcast links and directly modulated DFB node area segmentation on a single fiber. In combination with CWDM (15 wavelengths The authors wish to express deep after exclusion of water peak and 12 gratitude to Shamino Wang, Samuel Chang wavelengths after exclusion of CWDM and Ricardo Villa of Aurora Networks, who windows occupied by LcWDM™ spent hundreds of hours testing all aspects wavelengths traveling on the same fiber) and and components of the LcWDM™ systems. digital TDM technology in upstream, Without their dedication, the systems would LcWDM™ systems can support 6 to 8 have taken much longer to develop since its forward segments and 24 upstream segments conception in May of 2006. The authors also on a single fiber. This allows not only for appreciate the support of the Aurora segmentation of the area but also for node operations team in building countless addition in the growth areas without the need iterations of the system components. for fiber construction on the existing fiber Extreme gratitude also goes to Wim Mostert routes. In fact, it allows for fiber recovery to who first coined the idea in casual support commercial and other services and conversation on the way between the Aurora generate incremental revenue. headquarters and the company rental apartment. After peeling the layers and 40 ? Wavelengths DWDM analyzing all possible pitfalls, we realized we have a jewel on our hands, technology that benefits the entire industry.

1 4 to 8 ? ITU-T G.694.1 (2002), Spectral Grids for WDM LcWDM™ Applications: DWDM Frequency Grid 2 ? 2 CWDM Range ITU-T G.694.2 (2002), Spectral Grids for WDM 15km 30km 100km Applications: CWDM Wavelength Grid Figure 14: Capacity/Distance Limits of 3 T. Werner and O. J. Sniezko, Exploiting HFC Various WDM Technologies for Bandwidth Capacity To Compete With FTTH, 2006 NCTA Technical Papers Downstream HFC Optical Transport 4 M. R. Phillips, T. E. Darcie, D. Marcuse, G. E. Bodeep, and N. J. Frigo, Nonlinear Distortion A critical advantage of the technology Generated by Dispersive Transmission of Chirped described in this paper is its simplicity. The Intensity –Modulated Signals, IEEE Photonics Technology Letters, Vol. 3, No. 5, May 1991 transmitters carry entire forward load from 5 54 to 1002 MHz on the same wavelength and M. R. Phillips and D. M. Ott, Crosstalk Due to Optical Fiber Nonlinearities, Journal of Lightwave in this sense are equivalent in operation to Technology, Vol. 17, No. 10, October 1999 1310 nm directly modulated DFB in WDM CATV Lightwave Systems transmitters. The LcWDM™ system design 6 R. W. Tkach, A. R. Chraplyvy, Fabrizio follows the same rules as the design of a Forghieri, A. H. Gnauck, and R. M. Derosier, Four- system with 1310 nm DFB lasers, taking into Photon Mixing and High-speed WDM Systems, Journal of Lightwave Technology, Vol. 13, No. 5, account the required input level to the node May 1995 and power budget of the link (including fiber 7 D. Pike, The Effects of Reflections, 1989 NCTA and passive component losses). After Technical Papers 8 accounting for these design rules, the 30 km H. A. Blauvelt, P. C. Chen, N. Kwong, I. Ury, fiber length limit is rarely reached if more Optimum Laser Chirp Range for AM Video Transmission, 1992 NCTA Technical Papers than two wavelengths are used for downstream transport.

ACKNOWLEDGMENTS

Making FTTH Compatible with HFC James O. “Jim” Farmer Wave7 Optics, Inc.

Abstract competitive with FTTH networks that are being installed now, so we recommend that cable The evolution from Hybrid Fiber-Coax operators seriously consider installing FTTH for (HFC) to Fiber-to-the-Home is explored. The Greenfield situations, where they must install similarities between the technologies are something anyway. We shall show that it is noted, and we show how an operator who has easy to operate systems that are partially HFC been deploying HFC can start to deploy and partially FTTH. FTTH where it makes sense to do so, without excessive investment. WHAT IS FTTH?

INTRODUCTION You can think of FTTH as the logical extension of HFC networks. Cable operators for Cable operators are starting to deploy fiber- years have been pushing fiber closer and closer to-the-home (FTTH) in Greenfields, as well as to the home, to enjoy the benefits of greater for business applications. The preferred quality and reliability, and lower maintenance deployment strategies employ one of two costs. FTTH represents the end game, where standards for FTTH: EPON or GPON. This fiber has been pushed down to a node size of paper applies to both. Since the FTTH one home. When fiber is pushed this far, some deployments are small compared with the HFC interesting changes to the way the fiber is used deployments, it is important that minimal or no are possible. In FTTH systems: changes to the operator’s processes or headend technology are necessary in order to integrate -One fiber carries downstream and upstream the new with the existing. signals, minimizing splicing/connection costs.

This paper will review recent experience in -Data is carried on separate wavelengths from integrating FTTH into HFC systems. We’ll RF, resulting in incredible data bandwidths cover such topics as data management, video in without having to sacrifice any video spectrum. the two systems, and how voice is handled. Since data is carried on separate wavelengths There are certain architectural features of FTTH from RF, RF spectrum is freed up for more systems that can ease the initial cost of video. There is no tradeoff between data deployment. These features will be explored, bandwidth and the bandwidth available for showing how to make as much of the video. The entire RF bandwidth from 54 to deployment cost as possible be based on 1,000 MHz is available for video. success. -Reliability is enhanced due to the complete Cable TV HFC networks are much better lack of coax in the plant. There are no equipped to provide all modern residential corrosion issues, no mechanical pull-out issues, telecommunications services than are the no cracked shields, no sheath currents, etc. telephone company’s twisted pair networks. Therefore, we expect that cable operators will -While we are not going to advise you on continue to use their HFC plants for some time legal matters, we note that the maximum RF where they exist now. However, HFC is not signal level occurs at the side of the home, and is typically less than 20 dBmV. Thus, we would -Systems come with integral element not expect composite leakage index management – it is not an extra-cost add-on. requirements to exist. -With the exception of the set top box, all -There are no amplifiers to balance: the equipment is normally located outside the home, response you get at the headend is the response where the operator has access. The outside you get at the home. Always. equipment is normally powered from the home (with battery backup for lifeline voice), -As shown below, the manner in which minimizing the operator’s electric bill. upstream signals are handled results in no upstream noise funneling. -The FTTH network is architected to carry IPTV should the need arise. IPTV and broadcast -Since the plant is all-dielectric, there is no video may be carried at the same time. ingress into the plant. TYPICAL FTTH ARCHITECTURE -Typically there are no actives in the field, and no power connections. Reliability is Figure 1 illustrates a typical FTTH increased and maintenance is reduced. physical architecture. The video headend is just as it is for HFC. The voice facility is not -Service disconnect is an integral part of changed from HFC – FTTH networks can many FTTH systems, removing the need for support the same protocols specified by truck rolls when a subscriber disconnects or CableLabs. The switch(s) used to interconnect connects. data in the headend are the same: Gigabit Ethernet links are usually used to connect to the -Expensive CMTSs are not needed in most FTTH equipment. No CMTS is needed for the cases. Data connections are usually gigabit FTTH portion. A new unit, called an Optical Ethernet to your headend switch. Line Terminal (OLT) is used as the data interface with the FTTH plant. Logically, you can think of it taking the place of the CMTS, though the analogy is not precise. A typical OLT is pictured, and serves up to 2300

Figure 1. Typical FTTH Architecture subscribers in a typical configuration, in under home, and taking power from a special ring 1/3 of a standard rack. That gives you up to installed on the power meter. Power can be 6,900 subscribers served per rack. taken either before or after the meter. If lifeline voice service is provided, a battery is used to The video headend drives an externally provide a minimum of 8 hours standby/2 hour modulated 1550 nm optical transmitter, whose talk time. The battery condition may be output is split and amplified (not shown) as monitored through the EMS. necessary to supply signals to all subscribers. Only one transmitter is normally needed. The Other configurations of ONTs may be output of the 1550 nm transmitter is wave available. For example, a video-only ONT may division multiplexed (WDM’ed) with the data be available for customers who take basic video carriers handled by the OLT. Standard only. For business applications, there may be wavelengths used in FTTH systems are small, indoor-only ONTs that provide data ports downstream data on 1490 nm, upstream data on only. Some ONTs provide only one data port 1310 nm, and video on 1550 nm (any suitable and one voice port. Other ONTs provide standard ITU wavelength). multiple ports. Finally, ONTs designed for multiple dwelling units (MDUs) may be Normally the network to the home is all available, with ports to serve several apartments passive, consisting only of single mode fiber from one ONT. All ports are controlled through optic cable of the type cable systems normally the EMS. use, and passive splitting. Distributed splitting (taps) can be used, but it is more common to use THE ONT RF PORTION concentrated splitting, where all splitting is done in one or two locations. The network Figure 2 illustrates the ONT with some between the headend and the home is called a optional features included. The single fiber passive optical network, or PON. While it is from the PON enters the ONT at an optical technically feasible to serve 64 subscribers per wave division multiplexer (WDM). The top PON, serving 32 subscribers per PON is more port of the WDM passes the 1550 nm broadcast common. The cost differential, while not zero, signal to a broadcast RF receiver not unlike the is not great. receiver portion of an HFC node. The output level is usually set to allow several TVs to be The termination at the home is called an connected. Levels of +12-18 dBmV are optical network termination, or ONT. A common. Recall that the ONT is mounted on number of different ONTs are available, but the the house, so this level exists at the side of the workhorse configuration is a box that is home, not at the pole. AGC is often used to mounted on the side of the home. It has an RF compensate for optical level variation. The output for video, one or more data ports AGC may sense the RF signals level, but in this (10/100Base-T), and one or more voice ports application it is more satisfactory to sense the (RJ-11 with terminal block). RF return support optical level and correct the RF level. for set top upstream communications is available from some vendors. All ports are If RF return to support set tops is used, controlled through the element management the RF diplexer separates the upstream signal, system (EMS), and may be remotely turned on digitizes it, and sends it to the headend, where a and off. The ONT is usually powered from a special device is used to reconstruct the RF to power supply mounted in the home. supply it to the set top control system. There are Configurations are available that allow some variations in which the signal is mounting the power supply outside of the demodulated at the ONT and the demodulated packet is transmitted to the control system. No provides about 37 Mb/s downstream and data is transmitted except when there is a usually about 8 Mb/s upstream, spread over transmission from a set top in the home. however many subscribers, maybe 100 or so, if i Typically the threshold for determining when there is a typical application. The upstream data data is present is set toward the high end of the rate is 1 Gb/s, so you can offer outstanding range of which the set top is capable upstream bandwidth as well as downstream. (accounting for passive splitting/combining in This works especially well for gaming and peer the house), to minimize the effect of any noise to-peer applications. Following the data in the home. When a set top transmits, usually transceiver is the digital processing section. The one or a very few data packets are sent through front-end is a protocol-specific chip that the data portion of the FTTH system. You can manages data transmission according to either see that this method of transmitting the the EPON or GPON standard. The digital upstream RF precludes noise funneling in the processing back-end provides one or more upstream direction. If a home has a noise 10/100BaseT connections for data, and also problem it would be easy to identify based on includes media conversion to support standard the IP address of the transmission. analog phone lines.

DATA PORTION In Greenfield applications many homes are now being built with category 5 cable so that you The bottom port of the WDM (Figure can connect the cable directly to the ONT. If 2) passes the 1490 nm downstream data signal desired, a standard cable modem gateway is to an optical transceiver, and accepts from the connected to provide a firewall and transceiver the 1310 nm upstream optical signal. Data modulation is done at baseband (on-off keying) of the optical transmitter. The data rate at the transceiver is 1 Gb/s or higher, spread over only 32 subscribers. Compare this to DOCSIS, which on a per-RF-channel

Figure 2. Optical Network Terminal (ONT) DHCP services for the home. For Following standard practice, the downstream existing construction, there are several options. signals to the HFC plant and the upstream Many operators are installing cat5 cabling, but signals from that plant are transported to and some are opting for one of the standards from the node on separate fibers. lower available for putting data over coax. These frequency channel amplitude by about 4 dB at standards are HPNA over coax, and MoCA. channel 2, decreasing the boost at higher More on those options later. frequencies, to the point of little boost above about 200 MHz. The reason for this boost is MORE ON THE HEADEND that fiber optic cable exhibits a nonlinearity called Stimulated Raman Scattering (SRS) that Figure 3 illustrates the headend in more causes the downstream data, carried at 1490 nm detail. The headend shown is feeding both HFC (in accord with all modern standards), to plant (toward the center right) and FTTH at the crosstalk into the video channel, reducing bottom. For illustrative purposes, we have carrier-to-noise ratio (C/N). shown different VOD zones for HFC and FTTH, but of course this will vary depending on local needs. At the top is the headend switching/routing facility connected to the voice facility and other data services. It is connected to the CMTS for the purpose of data transmission on the cable plant, including voice.

Figure 3. Headend serving both HFC and FTTH Networks

The effect is frequency-dependent, that uses the upstream data on 1310 nm for affecting lower frequencies the most. By upstream transmission. That leaves the RF boosting the low-frequency channels, the C/N return from set tops that we need to handle. can be maintained equal to or better than that delivered by short HFC amplifier cascades. Figure 2 illustrates how RF return is handled at the ONT, and figure 3 shows RF DATA FEED IN THE FTTH PLANT return handled at the headend. Referring to Figure 2, RF upstream signals from set tops are The CMTS does not act as the data separated in the diplex filter in the ONT. The interface for the FTTH network, because data is RF return signal is digitized in the analog-to- carried as on-off modulation of the data digital converter (A/D). Processing consists of transceiver at different wavelengths than that doing some fast data compression and putting used for video. Thus, data doesn’t take the resulting digitized data into an IP packet. spectrum away from video. The interface taking The packet includes the IP address of the the place of the CMTS is called an Optical Line reconstruction hardware at the headend. This Terminal (OLT). It may or may not include packet is transmitted to the headend along with front-end data concentration (switching). It all other data. Note that the only time interfaces to the headend switch, usually using bandwidth is consumed for RF upstream either optical or electrical Gigabit Ethernet transmission is when a packet is actually (GBE) connections. available for transmission. This only happens when there has been a burst of RF from the set RF RETURN IN THE FTTH PLANT top converter.

In HFC plant, we handle RF return by Figure 3 includes the RF reconstruction. providing an upstream path between 5 and Upstream data from the FTTH plant appears at about 42 MHZ, using diplex filters at the input the GBE connections at the ONT, which are and output of each amplifier and at the coax connected to the headend switch or router. The side of a node. We normally use a second fiber headend switch routes the RF return packets to to return RF signals to the headend from an a special-purpose headend product that upstream RF optical transmitter in the node. reconstructs the RF signal. The reconstructed FTTH doesn’t have an RF return path for RF return signal is combined with RF upstream signals needing it. Some have proposed adding signals from the HFC plant if desired, and the a modulated laser at each home to combined signals are supplied to the receiver of accommodate RF return, but this is less than the set top control system. Since the destination satisfactory from cost and maintenance for the RF return signals is determined by the IP standpoints. address at the ONT, the RF reconstruction can be located anywhere in the plant. It is only The normal applications for the necessary to assign an IP address to the upstream plant are upstream data from DOCSIS reconstruction unit at the headend. Then part of modems, RF return from set tops, and perhaps the provisioning of each ONT is to tell it the IP status monitoring. In some instances there will address of the RF return packets. They will be contribution video from the field being arrive properly at the reconstruction unit. returned to the headend to be turned around. In the FTTH plant, we don’t need the upstream for The RF return system as shown can DOCSIS (except for DSG – more on that later), handle any of the standards in widespread use and status monitoring is replaced with a today, including SCTE 55-1, SCTE 552, and complete element management system (EMS) DSG (DOCSIS Set Top Gateway). Somewhat different provisions must be made for each, the plant, and is followed by a short cascade of due to the properties of the different standards. amplifiers. In the PON, the 32 (typical) Consult with the manufacturer of your ONTs subscriber PON consists of an all-passive for specifics. network, with the conversion to electrical being done on the side of the consumers’ homes. All Because of the way RF return works in that is in the field is optical fiber and passive FTTH plant, there is no noise funneling. splitters, until we get to the home. Referring again to Figure 2, the only time there is a transmission from the RF return section is Note that each PON has a WDM after it, when an upstream transmission is detected. The which combines the broadcast video at 1550 nm threshold for detection is set as high as possible, with the data at 1490 nm (downstream) and consistent with the need to make sure all set 1310 nm (upstream). Since each PON has a tops can reach that threshold within their WDM, it is obviously possible to further divide specified maximum output level, taking into the RF signals to get more frequency reuse in account in-home splitting and cable loss. The the FTTH plant, just as you do in the HFC packet can only be transmitted when nothing plant. However, if you wish to take advantage else in that PON is being transmitted, and of it (your option), you have additional though the RF reconstruction hardware in the spectrum available in the FTTH plant that you headend (Figure 3) services many PONs, only don‘t have in the HFC plant. You are not one signal can be passed to it at a time. Thus, loosing spectrum for data and voice, since they the system inherently offers protection against are carried on separate wavelengths. And you noise funneling. In the unlikely event that you probably have more spectrum available. The have one home that is generating enough noise RF FTTH spectrum extends to at least 870 to make the RF return circuitry think a signal is MHz, and 1,000 MHz really works (it is often present, then that noise cannot affect reception not specified that way for detailed technical from any other home. Furthermore, the reasons, but it works). So you may avoid as offending home is easy to locate: you simply much frequency reuse as youi need in some sniff upstream packets and note the source IP HFC situations. address the illegal packets are coming from. Also note that for SCTE 55-1 systems using CURRENT VIDEO PRACTICE contention signaling, the system can actually improve efficiency by queuing multiple Entities deploying PONs today have a transmissions. number of philosophies regarding how to deploy video. While some are deploying IPTV PON Vs. NODE exclusively, we find that most (including Verizon) are deploying broadcast video, while The OLT shown is typical in that it reserving the option to deploy IPTV in the includes a number of optical interfaces, each future. The reasons for the continuing serving an individual set of subscribers. That popularity of broadcast video, especially in set is referred to as a PON, or passive optical competitive situations, are several. Broadcast network. A PON in this context may be video is more mature, as the cable TV industry considered as analogous to a node in HFC plant. knows well. There are many more features In FTTH, a PON is effectively a node serving available on broadcast set tops than there are on typically 32 subscribers per node. The node IPTV set tops, and the user is accustomed to the analogy breaks down though, when we consider broadcast experience. Furthermore, IPTV does what is in the field. In a node, the node necessitate certain in-home wiring (or use of the (conversion between optical and electrical) is in devices discussed in the next section) not needed for broadcast video. Finally, the ability on in-home phone networks, but it was quickly to supply analog-only services without a set top realized that the same standard would work well where that is the only service desired, is a over coax. The frequency band occupied is 4-21 subscriber convenience and an operator cost MHz, so it overlaps spectrum often used for RF savings. return. Equipment supporting HPNA is readily available from a number of vendors. A likely future scenario is to provide for iii a broadcast basic service (analog-only or analog The other standard, MoCA, occupies and digital) on broadcast, while offering video- spectrum above 870 MHz, so it is compatible on-demand (VOD) and similar services on IP. with RF return. Use of that higher spectrum This gives you the best of both worlds: raises concern about it’s viability in older broadcast where it is most efficient, and IPTV home wiring, but the MoCA association where it is most efficient. While we are not reports good throughput in nearly all homes aware of any operators who have implemented tested. this scenario yet, we do know of operators who are expecting to implement it in the next few Both standards are used in essentially years. the same way. A unit at the ONT, usually designated as the master, transmits data presented to it on one of the 10/100Base-T IN-HOME NETWORKING connections at the ONT. An adaptor (slave or client) at each computer converts the RF signal Figure 4 illustrates some in-home back to a 10/100 connection. These adaptors are networking options that are available for use often called dongles. with FTTH and, with suitable modification, with HFC networks. To the left are shown the The two data-over-coax standards can ONT RF and data connections. Since most carry IPTV data to set tops and can also be used homes in North America are wired for coax but for data delivery and in-home networking. not data, there has been a lot of work on putting Figure 4 shows client-to-client data flow. The data on the coax network in the home. The standards rely on limited isolation between cable industry has been following this work splitter ports, plus reflections, to achieve client- closely. Two standards have emerged, and the to-client data communications. This usually selection between them may be coming down to works well, as the standards have adequate whether or not cable TV set tops are being used. dynamic range to work over a wide range of Measured throughput for both standards is on attenuation. If a problem were to arise, you can the order of 100 Mb/s. Both standards support artificially worsen the return loss in the band of data delivery to all computers in the home over interest by installing a filter with a stop band at coax, networking of the computers, and the device operating range, at the RF output of delivery of IPTV if and when desired. the ONT.

HPNA 3.0/3.1 can be used where RF The use of HPNA or MoCA for in- return from set tops is not needed, such as home networking as well as for delivery of where IPTV is being employed for premium IPTV (if both are being done) will demand that services. Originally HPNA was an acronym for quality of service (QoS) be applied. Both Home Phone Networking Alliance, but the standards have the ability to do basic QoS organization changed its name to the HPNA functions, at least as far as prioritizing one type ii Alliance. This standard was developed for use of traffic over another based on 802.3 quality markings.

Figure 4. In-Home Networking

DIGITAL TURN-AROUND VOICE COMPATIBILITY

Digital turn-around, where you use the Voice compatibility between FTTH and upstream data path to send video back to the HFC is easy, and there are systems using the headend to turn it around and send it to same soft switch for both today. Many FTTH subscribers, is easy in FTTH. This is being done systems support MGCP/NCS protocol, and they today, where systems are using their facilities to can support SIP. The CableLabs specified NCS carry local sports. They convert the video to is a profile of MGCP (some people say it the IPTV at the venue, give the packets the IP other way around), so it is common to support address of the turn-around device at the both signaling standards. SIP is starting to gain headend (or any other point in the network), add a lot of traction in the industry, both in HFC quality of service (QoS) parameters, and the networks and in FTTH networks. turn-around happens. No worries about ingress, A difference between the two networks no issues of finding spectrum for the upstream is the way QoS is implemented. DOCSIS uses video, no plant qualification, no headend re- a special QoS paradigm in which a cabling. Anywhere you have an ONT in the communications path is set up at the beginning plant is an injection point for entertainment- of a call and torn down at the end of the call. quality video you can turn around to your The soft switch (or other voice facility) is subscribers. responsible for initiating set-up and tear-down of the circuit. FTTH systems, on the other hand, use IETF- and Ethernet-standardized prioritization. The advantage is that there is no their new subdivisions, because they have need to set up and tear down the connection. learned that FTTH adds value to the homes they Voice packets are identified and are transported build. Cable TV operators are in a unique with high priority. Soft switches don’t have any position to service this business, and at the same problems distinguishing between calls placed time, equip themselves for a future in which on FTTH and on HFC, and there are people they will need FTTH in order to compete. today who are operating mixed systems, using the same switch for both HFC and FTTH. END NOTES

i CONCLUSION DOCSIS 3.0 provides more bandwidth by bonding channels, but this takes channels that cannot then be used for video. It is very easy to add FTTH to an HFC ii http://www.homepna.org/ network. This is advisable where you have iii http://www.mocalliance.org/ greenfield opportunities close to HFC plant. Many land developers are demanding FTTH in

MULTI-WAVELENGTH ACCESS NETWORKS: A PRACTICAL GUIDE TO IMPLEMENTATION Ray Thomas, Time Warner Cable Venk Mutalik, C-COR

Abstract the forward path and a second fiber was used for the return path. Some cable operators Continued growth in residential data, built out networks using fiber sheaths voice and video traffic is a success story that containing four or six strands of glass to each has driven MSOs to increase throughput optical node, while other cable operators used capability in all parts of their networks. as few as two or occasionally more than six. When increasing throughput capacity, In the cases where six fibers inside a sheath different parts of the end-to-end network are going to a specific area of the system, 3 require different approaches. In the access optical nodes could be served. If a fourth (or “distribution”) part of the network there node was needed to serve increased data have been advances in optical technology traffic in that area, another pair of fibers that lend themselves to being a rapidly would be required. A typical solution is to deployable, robust, cost effective solution. install additional fiber to serve that area. The Multi-Wavelength Access Networks are an costs for constructing fiber add up on a per- additional tool in the cable operator’s foot or per-mile basis. The process of toolbox that will allow MSOs to increase obtaining construction permits and working throughput capacity or provide additional during bad weather conditions can add delays services on their existing infrastructure at to construction projects and the attending significantly less cost than new fiber service disruptions may also have additional construction. negative consequences.

This paper will lay out a practical guide However, the development of multi- to implementation of multi-wavelength access wavelength optical technology presents the networks. It will first present a model that opportunity to make increased use of existing explains the various benefits and trade-offs of fiber and delay the need for building out implementing Multi-Wavelength Access additional fiber. Additional forward and Networks. The paper will then document all return paths can be added to the fiber already the optical impairments that accrue within in place. Construction delays along busy access networks, and conclude with highways and streets or in back easements recommendations for the practical with difficult access can be postponed. The implementation of multi-wavelength systems. use of multi-wavelength technology allows for additional throughput capacity to be BUSINESS MODEL FOR turned up quickly. IMPLEMENTING MULTI-WAVELENGTH ACCESS NETWORKS A comparison of the relative costs of multi-wavelength optics versus construction One of the driving forces for the use of provides a compelling justification. Each multi-wavelength optics in the access mile of aerial fiber construction costs in the network is the requirement to increase range of $12,000-$16,000 (depending on a throughput capacity. Optoelectronics handful of factors). Constructing 10 miles of equipment, installed over the past 15 years, aerial fiber would therefore cost $120,000 or employed a design in which one fiber carried more. Underground fiber construction can cost up to twice as much per mile as aerial services segment is a market in which cable construction. On the other hand, the total operators have staked out a solid business. cost of multi-wavelength optical equipment Businesses represent a market with strong for both ends of the fiber run is less than the growth potential for cable operators. Cable cost of constructing two miles of new fiber. operators have a range of options for Chart 1 shows the relationship between the providing service to small, medium, and costs of construction versus multi- large businesses. A few of these options wavelength implementation. This is a include: powerful financial incentive for the use of multi-wavelength optics. o a direct fiber feed into the business

o cable modem business class service

o wireless DOCSIS

A widely preferred approach is to use fiber for serving businesses. Although other options such as wireless DOCSIS can be used as a temporary solution to quickly establish service by reaching across obstacles (railroad tracks, large parking lots, rivers), fiber has long been established as a reliable solution. Chart 1 – Multi-wavelength Access networks are much more cost-effective than construction Most networks were built with spare fiber Existing 1310 Multi- capacity in the access part of the network. design wavelength As fiber is used to connect new business design customers, the spare capacity in the fiber Construction of If fiber exhausted Only required if additional fiber must build extending fiber a sheaths is reduced. Multi-wavelength additional fiber short distance to technology is an alternative path to providing (overlash aerial or serve additional trench in u/g) node(s) additional services on the networks without Time required for Requires weeks Can be installed incurring the cost of new fiber construction. implementation to design, plan, in under a week obtain any or even in one required permits, night THE FIBER SPECTRUM and construct fiber Costs Approx $12k/mile Cumulative The CWDM wavelength plan is detailed aerial or $30k/mile differential cost in ITU Recommendation G.695, which was u/g for additional wavelengths is ratified in January 2005. The plan provides much less than for 18 wavelengths spaced 20 nm apart over 10% of fiber construction cost a range from 1271 to 1611 nm. Typically the Table 1 – Multi-wavelength Access networks can be 1371 and 1391 nm wavelengths are the implemented more quickly than construction designated water peak wavelengths of deployed optical fibers (Figure 1 provides Continued growth in residential services insertion loss for a 20 km fiber link for fiber drives the need to continue increasing with and without the water peak). throughput capacity. The commercial Cable operators can choose to reserve the compared to long haul transport networks of 1531 and 1551 nm bands for possible around 500 to 1000 km - there still are deployment of services utilizing the DWDM numerous optical impairments that could spectrum. Long haul 1550nm optics, EDFA, cause measurable degradation of the RF and QAM overlay architectures utilize this spectrum. The ability to identify all such portion of the optical spectrum. The wide impairments and manage their impact on the CWDM channel spacing allows the use of RF spectrum is central to making multi- lower cost uncooled DFB laser technology wavelength access systems work. for reverse path transmitters. Optical impairments are artifacts in fiber Additionally, CWDM channel spacing networks and the fiber itself that impacts how enables the use of cost effective, well the RF spectrum is carried in the environmentally hardened optical passives network. There are two broad classes of for field deployment. These passives can impairment: linear and non-linear. It is typically be obtained off-the-shelf from generally the case that non-linear suppliers and do not require unique impairments are dependent on optical specification considerations. The CWDM intensity whereas linear impairments are not. specifications for active devices promote The two classes of impairments can be GbE SFPs, reverse path analog transmitters further divided into single and multiple and forward transmitters. The CWDM spec wavelength non-linearities for the optical therefore provides for a significant increase non-linear impairments, and fiber linear in fiber capacity and promotes a level of plug effects and optical passive effects for the and play capability. optical linear impairments.

Figure 1 – Some wavelengths are more suitable for Figure 2 – Identifying Optical Impairments specific services than others. Linear impairments, like dispersion and From Figure 1, it is also seen that the familiar passband ripple, impact RF spectrum figures DWDM wavelengths are in the 1525 to 1560 of merit, such as CSO and CTB. Most often, nm region and are covered by the spectrum linear impairments are the performance of the 1531 to 1551 nm CWDM bands. manifestations of optical passives/fiber design and manufacture. Careful attention to OPTICAL IMPAIRMENTS WITHIN (and verification of) product specifications ACCESS NETWORKS can limit the effects of this class of impairment. Although distances involved in access networks are quite modest - around 20 km as Single-wavelength optical non-linearities decreases becoming essentially extinct after include the well known Stimulated Brillouin about 200 nm, the 4WM potentially flares up Scattering (SBS) and the lesser known Self- as the wavelength spacing decreases and one Phase Modulation (SPM). The SBS is often of the signal wavelengths approaches the compensated for by manipulating the optical fiber dispersion zero point. spectrum and/or limiting optical launch power in the network. Typical CWDM multi-wavelength systems employ two forward wavelengths Multiple wavelength non-linear spaced 20 nm apart. Here, the SRS is the impairments, such as Stimulated Raman dominant non-linearity and the system reach Scattering (SRS) and Cross Phase is limited by the total power launched into Modulation (XPM) induce crosstalk between the optical network. It is possible to reduce two or more wavelengths. Crosstalk could be the SRS effects by reducing the wavelength between wavelengths carrying similar spacing below the standard 20 nm used in services, such as two wavelengths carrying CWDM and thereby increase the power signals in the reverse direction; two launched into the network and/or increase the wavelengths carrying GbE; or two number of forward wavelengths. However, wavelengths carrying forward signals. In this approach can also substantially increase these cases, the RF spectra coincide and the the 4WM potential when additional variables crosstalk results in a direct impact on the RF such as dispersion are introduced in the performance. system, sometimes leading to a less robust system. A nonstandard wavelength spacing Crosstalk may also crop up between plan could also increase the system cost due wavelengths carrying dissimilar services, to lower volumes. such as a forward wavelength and a reverse wavelength, or a forward wavelength and a The ability to adequately model and test GbE wavelength. In these cases, if the all aspects of fiber impairments with frequency spectra overlap, there could be a particular emphasis on the number of direct impact on the RF performance. For wavelengths, polarization and fiber example, since the GbE and forward RF dispersion, along with the earlier mentioned spectra overlap, GbE signals (such as variables such as maximum power launch, spurious spikes when the GbE is unloaded) wavelength spacing and optimally selected could bleed through from the GbE RF optical filters, is critical to promoting a spectrum into the forward spectrum due to robust cost-effective solution that also fiber non-linearities, causing measurable satisfies capacity needs. performance degradation. DESIGN TRADE-OFFS For access network design, SRS and 4 Wave Mixing (4WM) are the dominant non- The analog realm features familiar RF linearities. Both of these are sensitive to trade-offs, such as the fact that CNR can be polarization and become progressively worse traded off to achieve better CSO and CTB with higher launch power. Both these and vice-versa. Similarly, multiple phenomena depend upon the wavelength wavelength access networks will have trade- spacing and fiber dispersion as well. While offs to make sure that each wavelength SRS becomes worse with increasing spacing passes through the network without up to approximately 100 nm and then impacting the other wavelengths on that wavelength network can still be enhanced by network. supporting lower optical node receive power. It is often the case that adding a node results in a shorter RF cascade with inherently less CNR degradation, so lower optical receive power can be used without degrading the end-of-line performance. For this reason, segmentable nodes, placed deeper into the network, are particularly well suited for multi-wavelength access networks. Figure 3 – Multi-wavelength network system performance requires adequate single wavelength WAVELENGTH PLAN performance and limited optical interference

The service disruptions that plague new To employ these trade-offs effectively, the fiber construction can be minimized for inherent optical linear and non-linear multi-wavelength access networks if a impairments should be studied to identify and wavelength plan is considered in advance of quantify their impact on the overall system. the design. This plan should ideally proceed

sequentially from an examination of fiber A good deployment strategy would link lengths and fibers available for include comprehensive testing and analysis deployment (fiber link description) to the of all optical parameters of the system so that services intended for each fiber (deployment design rules for field deployment can be package description) and further into future devised. These rules may govern the plans for services and fiber usage. Important locations of optical wavelengths such that the aspects of future use planning include overall optical crosstalk is minimized. consideration for 10 GbE usage, preservation

of the DWDM band and the desire for route Another rule would consider appropriate redundancy. intermixing of wavelengths of diverse RF

spectra to ensure that the optical level of the RECOMMENDATIONS FOR PRACTICAL composite signal being launched into the IMPLEMENTATION fiber remains below specified limits.

The following well tested design Another useful strategy consists of examples present architectures ranging from identifying and investing in optical passives simple to complex that progressively allow that support the selected wavelengths and higher and more effective utilization of have adequate isolation and loss installed optical fiber. The fiber utility table specifications. These will often be unique to at the bottom center of the figures will keep a a specific application. For optimal economic running tally of the number of wavelengths efficiency, it is good practice to set a standard used in each fiber. Although the architectures usage plan for wavelengths in order to drive are presented in the context of the CWDM higher volumes of identical optical passive standard, similar architectures can be configurations. conceived for DWDM or other multiple

wavelength allocation plans. Since the launched power of a multi- wavelength system is limited by optical non- linearities, the fiber reach of a multi- The classic architecture in Figure 4 is essentially characterized by a fiber pair from the hub or the headend terminated into a node. Each fiber then carries only one wavelength, generally at 1310 nm.

Figure 6 – CWDM in the Forward and Reverse

Figure 4 – Basic Architecture Figure 7 carries the previous architecture further. Here the operator may add the Figure 5 illustrates a very cost effective Gigabit Ethernet (GbE) traffic to the way of increasing fiber utility and providing previously analog/QAM HFC plant. reverse segmentation capability. This architecture maintains the forward and reverse designations on the available fibers and is minimally invasive. Higher utilization of the fiber is possible when the two fibers are collapsed into one. That strategy is the subject of the next architecture.

Figure 5 – CWDM in the Reverse

Figure 6 represents an improvement over the previous architecture in that the node can now be segmented in the forward and reverse path. This architecture enables the operator to have specific fibers designated for forward or reverse purposes and is least disruptive in providing service augmentation. Please note however that the two wavelengths should have the same analog broadcast signals, but Figure 7 – GbE in the Forward and Reverse Fibers could have different QAM 256 narrowcast signals. Figure 8 shows the three different services propagated on a single fiber. An architecture of this type provides the most effective usage of the optical fiber. A single fiber is therefore able to provide 2-way segmentation of the forward narrowcast QAM 256 signals, 4-way Traffic continues to grow in cable segmentation of the reverse signals and 2 bi- operator networks. In some metro areas the directional GbE business service links. rate of growth has been astonishing. Development of multi-wavelength optics technology has progressed to the point where it offers a very attractive alternative to construction of additional fiber for increasing throughput capacity. This provides an opportunity for operators to save large amounts of capital by taking advantage of multi-wavelength optics.

REFERENCES

1. Gigabit IP and Analog/QAM Traffic Compatibility in HFC Networks - A case for Modern physical layer Figure 8 – Typical CWDM Capacity per Fiber convergence, Shamim Akhtar, Doug Weiss, Comcast Corp., Venk Mutalik, As indicated earlier, the RF input to each Zoran Maricevic, Marcel forward transmitter is independent; however, Schemmann, David Heisler, Don the analog broadcast content of the two Wesson, Liyan Zhang, C-COR Inc., transmitters should be the same. The QAM SCTE Conference on Emerging 256 narrowcast content for each of the Technologies, 2006 transmitters can differ. The actual capacity of the optical fiber is much higher than 2. Cable Television Technology, Second represented here however. The higher Edition, Chapter 13, Wavelength- capacity is obtained by employing additional Division Multiplexing, The Morgan wavelengths that are carefully chosen to be Kaufmann Series in Networking compatible with the system needs after 2004, by Walter Ciciora, James considering the optical impairment Farmer, David Large, Michael Adams mechanisms discussed earlier.

CONCLUSION 3. Nonlinear fiber optics, Govind P. Agrawal, Academic Press, Second Multi-wavelength systems in the metro, Edition, 1995 long haul and transport arena have been designed for many years now in the form of 4. Enhancing cable architectures with Multi-Wavelength optical networks. Multi- WDM - Grafting WDM onto existing wavelength systems are increasingly being cable systems, Venkatesh G. Mutalik, considered for use in the access (distribution) Communication Engineering and part of the network. The technology behind Design (CED Magazine), February multi-wavelength access networks has been 1997, Page 54, WDM Series evolving over the past year to offer significant new capability.

5. ITU-T Recommendation G.695 - Optical interfaces for coarse wavelength division multiplexing applications: Series G Transmission systems and media, digital systems and networks (01/2005)

6. DWDM: Matching Technology Advancements with Business Requirements, Venkatesh G. Mutalik, National Cable Telecommunications Association (NCTA) Proceedings 1999

7. Maximizing Today’s Network: Increasing HFC Network Capacity without Additional Fiber. Jeff Sauter and Bill Dawson. Communications Technology magazine, July 1, 2006.

CONTACTS

Ray Thomas, Principal Engineer, ATG Time Warner Cable Tel: 720-279-2729 Email: [email protected]

Venkatesh Mutalik, Chief Technologist C-COR, Inc. Tel: (203) 630-5763 Email: [email protected]

NARROWCAST SERVICES – UNIFYING ARCHITECTURES Glen Hardin, Time Warner Cable

Abstract programs from the digital tier through SDV is projected to save 50 percent of the bandwidth Cable advertising is a primary source of required when compared to broadcasting revenue for the cable industry. It is a massive through normal mechanisms. successful industry by its own right. The ability to target an advertisement to a subset Narrowcast Bandwidth is the primary of the cable plant or “ad zone” is one of the technology infrastructure that forms the key attributes that differentiates cable’s foundation to allow cable to deliver the advertising capability from that of the aforementioned services of the triple play; national broadcasters. These uniquely zoned Voice, Video and Data. The Narrowcast advertisements are now more valuable as they Bandwidth is native to the cable platform, is are targeted to their intended recipients and one of cable’s primary differentiators and narrowly broadcast to a subset of the cable serves as a springboard on which next plant to reach the recipients. generation services are launched. Cable’s current successes can be traced back to High Speed Data (HSD) is another highly investing in the Hybrid Fiber Coax (HFC) successful cable service that cable deployed architectures. Cable’s future success will be over 10 years ago and is still experiencing built upon the extension of the HFC network, strong growth. On a wide scale, Cable’s HSD the Narrowcast. It is key to building cable’s offering is leading the competition with sustainable network and is cable’s primary 8Mbps and 10 Mbps HSD offering. tool to address competition.

Closely tied to the HSD service is the This paper seeks to provide context for the Voice Over Internet Protocol (VOIP) phone Narrowcast - its origins, growth, future and service. This relatively new service has importance to cable’s future. quickly become another mission critical offering for the cable operator representing Defining “Narrowcast” new sources of growth and income. First, a few terms need to be defined to clarify There is little doubt that Video On Demand their use in this paper. (VOD) is a success. According to the industry press, across the industry, VOD revenues Broadcasting is the transmission of have grown to over a billion dollars of programs or signals to the entire cable plant. revenue for the cable industry per year. VOD That is to say, source signals (analog video, is widely deployed across all major markets digital video or data) are modulated onto a and is the cornerstone of the digital offering. frequency where it is broadcast across the cable plant unmolested or uninterrupted to the One of the next generation key entirety of end users. Cable was built on that technologies to unlock Cable’s bandwidth capability to transmit signals to the entire potential is the delivery of television through plant and it will always continue to broadcast. the Switched Digital Video (SDV) infrastructure. Broadcasting the select Narrowcast Source Integration - Analog, Digital & Data

GigE Backbone Narrowcast RF Narrowcast Residential QAMs Combining Service RF Distribution Groups 2000 QAMs SG-1 QAMsQAMs VOD Servers QAMs Homes Passed

QAMs SG-2 2000 QAMsQAMs Homes QAMs Passed Ad Servers 2000 QAMs SG-3 QAMs Homes QAMs QAMs Passed

2000 SG-4 Switched QAMsQAMs Homes QAMs Digital QAMs Passed Video

CMTSQAMs QAMs QAMs

RF Two-Way QAMsQAMs QAMs Residential Broadcast Signals

Narrowcasting is the narrow A Narrowcast Service Group is the “broadcasting” of source signals (analog subset of subscribers within a cable plant that video, digital video or data) to a subset of the is served by a unique set of source signals cable plant and not indiscriminately to the (analog video, digital video or data). plant on the whole. A Channel is defined as a single 6 In this manner, end users will receive MHz frequency slot, whereas a Program is unique source material based on their defined a single video source. To put it into geographic location within the plant. This is perspective, an analog source would occupy accomplished by inserting unique or targeted an entire 6 MHz channel as a single program source and modulating it onto a RF frequency stream, whereas a digital program stream may that only serves a small subset of the cable be only one of ten programs on a channel. A plant. These Narrowcast modulators deliver Stream is the transport of a program across a unique RF channels (bandwidth) and thus channel at defined bandwidth. In the context unique sources to a subset of the cable homes. of this paper, streams can be analog, VOD and SDV. It needs to be acknowledged that there is, in fact, both a Forward Narrowcast and a Return Narrowcast, but to avoid too much complexity in the discussion, this paper will focus on the forward Narrowcast.

Understanding the Architecture foundation for all of all the following services: Cable systems were originally constructed as purely broadcast networks • Two-way RF Communications using cascading amplifiers in series to distribute the broadcast signals to the entire • Advertising Zoning cable footprint. This architecture was interdependent and was as much an art as a • High Speed Data science to keep the distribution network in “balance” across all the amplifiers within the • Voice Over Internet Protocol network. • Video On Demand/Start Over The figure above represents the basic constructs of the Narrowcast architecture. • Switched Digital Video Modern Hybrid Fiber Coax (HFC) architectures were introduced to compensate Each service type may serve a unique for the limitation and complexity of managing subset of the cable plant. Thus, within a given series of cascading amplifiers. The reference cable plant there could be an almost infinite architecture was based on roughly 500 homes number of different service groups based on per node and 4 nodes or 2000 homes per laser. service types. This is complex enough in the abstract but in the plant, this represents As part of this transformation, real- countless numbers of physical devices, wires, time two-way Radio Frequency (RF) combiners and splitters located across a large communication was introduced to lay in the geographic area (and this is just the forward foundation for advanced services. path). Narrowcasting of the two-way RF communications addressed the problem of Traffic Modeling Narrowcast Services trying to communicate to a very large population of Set-Top-Boxes (STBs) by Traffic modeling is the exercise in breaking the down the cable plant into estimating the simultaneous use of limited subsections. The two-way RF communication resources by a given number of users. In equipment would narrowcast on the same context of cable’s Narrowcast services it is the frequency, but due to plant isolation they estimation of the amount of bandwidth at peak would not interfere with each other. In that required per service type to deliver that way, the same narrowcast frequency could be service. “re-used” by each sub-plant. To put this into well-known context, the Narrowcast In cable’s context: communication technologies that provide the two-way RF communications are Scientific Two-way RF Communications – it is Atlanta’s Quadrature Phase Shift Key the number of STBs that can be attached to a (QPSK) modulators and demodulators and the given QPSK Mod/Demod without having too Motorola’s Out of Band Modulator (OM) and many collisions that would make Return Path Demodulator (RPD). communication impossible.

This basic technique of plant segmentation Advertising Zoning – is the exception. and spectrum re-use for creation of Although it is a Narrowcast service it is Narrowcast Bandwidth serves as the multicast to all users within a zone with guaranteed non-contested bandwidth and, requirement. VOIP is the first DOCSIS therefore, is not bound by a traffic model service to really demand QoS. Not to overly constraint. understate the importance of VOIP QoS but it is by magnitudes less demanding than video High Speed Data – it is the modeling QoS. A phone call requires 128Kbps/sec of the number of cable modems that can versus 3.75Mbps for Standard Definition (SD) attach to a VOD sessions. Network architectures are CMTS port and provide the required being adjusted to compensate for this new bits/second service performance. QoS requirement on the data-networking infrastructure. Voice Over Internet Protocol Phone Service – it is the number of simultaneous Besides QoS, there is another phone calls that can be made at a given important concept to take into account when moment in time. performing traffic modeling on Narrowcast services; blocking. Blocking is the term used Video On Demand/StartOver - it is the to describe denial of service due to contention number of simultaneous VOD sessions that of resource. The familiar analogy is the busy can support at a given moment in time. signal associated with trying to make a call on Mother’s Day when everybody else is trying Switched Digital Video – it is the to do the same. There are not enough number of simultaneous switched television resources to fulfill everyone’s request at the programs that users can be watching at a same time. given moment in time. Either due to the QoS requirement for When analyzing the various traffic both VOD and VOIP each service is models of the different narrowcast services, it guaranteed its full bandwidth requirement for is important to note the key differentiator of that session or the session/call is denied or service type, and that is the Quality of Service blocked in its entirety. This is unlike “best (QoS) requirement. All video services such a effort” delivery of data services. VOD and SDV require near perfect QoS delivery, where as with data services like two- For each service type, there may be an way RF communication and data DOCSIS acceptable level of blocking. The Service services they have built-in mechanisms to Level Agreement for VOIP phones service overcome delivery issues such as collisions, and SDV services may require them to be dropped packets, burst or sporadic packet non-blocking where as VOD may be delivery and out-of-order packet delivery. considered a blocking service. Streaming video delivery does not have any of these mechanisms. If a video packet is The basic traffic model analysis dropped, arrives out-of-order, is improperly example will be performed on video service spliced into the main stream or is not properly due to the QoS requirement of video. paced in its delivery, a video artifact will be seen by the end user. Video delivery is all Video Narrowcast Modeling about QoS. Video Narrowcast utilization is based The exception to that statement is on three key factors, the number of VOIP telephone service. With the subscribers, the available bandwidth (i.e. introduction of VOIP, the data side of the streams both SD and HD) and peak network had to address this new QoS simultaneous utilization. In the examples only SD content is used to model for 45%, and markets have launched High simplicity. Definition VOD (HDVOD) that uses 15 Mbps instead of the SD rate of 3.75 Mbps. VOD & Narrowcast However, the biggest contributor to increase usage of the VOD Narrowcast is the StartOver Originally, the narrowcast for VOD service. was built to a 6% peak streaming utilization of digital households. Four RF channels were StartOver is an advanced VOD allocated with each channel supporting 10 technology that utilizes the existing VOD and VOD streams per QAM 256 for 40 streams Narrowcast infrastructure. Start Over enables per 666 Digital Households. the time shifting of live television without the need for upgrades to customer premises The basic math: equipment. Time-shifted television allows 500 Homes per node subscribers to re-start and begin watching 4 nodes per VOD Service Group their favorite broadcast TV program during 2000 Homes Passed per VOD Service Group any point in the broadcast window. Unlike Digital Penetration 33% (Digital Households) traditional on-demand services that have 4 QAMs (256) @ Payload of 37.5 Mbps license windows measured in days or weeks, 10 VOD streams per QAM at 3.75Mbps the Start Over content is only available to start a session within the actual broadcast window 2000 HP * 0.33 Digital = 666 Digital Homes of the particular content. 4 QAMs/SG * 10 streams=40 streams/SG With the launch of the StartOver 40 streams/SG / 666 Digital Homes = .06 service stream utilization within the or 6% narrowcast of service group has seen a 50% increase in utilization of the narrowcast Therefore, the 40 available streams are bandwidth. to be shared across the 666 Digital Homes within the VOD Narrowcast Service Group. Success is a high-class problem to solve and cable is well equipped to address VOD is considered a blocking service contention within the VOD Narrowcast. The where at peak utilization some requests are optimum technique to address contention accepted to be blocked denying service. within the narrowcast is to increase the Additionally, VOD is a uni-cast narrowcast number of channels per narrowcast service service. This means there is a one to one group. But this requires additional bandwidth relationship between narrowcast bandwidth that is typically not available. Therefore the use and user. primary tool at their disposal is what is termed “splitting” the VOD Service to As mentioned at the beginning of this “manufacture” enough bandwidth to support paper VOD is a huge success and with that the ever-increasing demand. success, careful management of the Narrowcast Bandwidth is critical to avoid block and denial of services. There are few major factors contributing to high use of VOD. On the system today there is more compelling content, the digital penetration has increased by ten percentage points to around Splitting of a Service Group - Before

Narrowcast RF Narrowcast Residential QAMs Combining Service RF Distribution Groups 2000 QAMsQAMs SG-1 QAMsQAMs Homes Passed

Splitting of a Service Group - After

Narrowcast RF Narrowcast Residential QAMs Combining Service RF Distribution Groups 1000 QAMsQAMs SG-1-A QAMsQAMs Homes Passed

QAMs SG-1-B 1000 QAMsQAMs Homes QAMs Passed

Manufacturing Bandwidth The VOD math - holding Digital Penetration In the previous example where the constant: VOD service group consisted of: 1000 HP * 0.33 Digital = 333 Digital Homes 500 Homes per node 4 QAMs/SG * 10 streams=40 streams/SG 4 nodes per VOD Service Group 2000 Homes Passed per VOD Service 40 streams/SG / 333 Digital Homes = .12 Group or 12%

To split a service group the number of The VOD math - adjusting Digital Penetration nodes defined in the service group are halved to 45%: thus a split service group now consists of: 1000 HP * 0.45 Digital = 450 Digital Homes 500 Homes per node 4 QAMs/SG * 10 streams=40 streams/SG 2 nodes per VOD Service Group 1000 Homes Passed per VOD Service 40 streams/SG / 450 Digital Homes = .089 or Group 8.9% Digital Penetration 33% (Digital Households) The drawing above illustrates the basic thresholds per service and an appropriate concept of splitting of the service group. response to provide additional narrowcast By splitting the SG-1 service group into SG- bandwidth is all that is required to ensure 1-A and SG-1-B the affect was to increase the optimum performance. This is why the term peak simultaneous streaming capacity from “manufacturing” bandwidth is a truism within 6 % to 12% and even when including the the cable industry. increased Digital Penetration of 45% the peak streaming capacity worked out to roughly 9%, Switched Digital Video & Narrowcast an increase of 3% per service group from before the split of the original service group. SDV works off the basic premise that not all digital programs are being watched in a Real work must be performed in a given service area at the same time, so why coordinated manner in the RF plant to make a broadcast them, why not narrowcast them just Service Group split occur without impacting like a VOD stream. the customer. On the physical level within the cable plant to accomplish a Service Group SDV can be thought as of moving the split, new QAMs are require to be: STB tuner out of the STB and placing it just in front of the narrowcast SDV QAMs. To • Installed provision these SDV Narrowcast QAMs, the digital programs are multicast onto the • Configured IP & Controller transport ring and when a client “tunes” a multicast join occurs on the edge device and • Wired into the combining and the signal is routed through the narrowcast distribution network (forward and QAM for distribution to the STB. reverse) When studying the narrowcast • RF balanced into the plant modeling for SDV the basis for modeling was changed from Digital Households to STB • All the STBs on that original service tuners. This was done to accommodate group must be notified that belong to devices that can have multiple tuners per the “new” service group device like DVRs. Where each tuner could be tuned to a different signal source at the same • Tested time. For every 250 to 500 STB tuners (DVR STBs have 2 tuners), a SDV narrowcast This work represents a manageable service group is created across the entire cable amount of work for the cable system and they footprint. have become very adept at managing the service group split to a science. The SDV math (more factors): 500 Homes per node This is the same basic technique used to 2 nodes per SDV Service Group address contention across any of the Digital Penetration 28.6% Narrowcast Services. Regardless of the Target size of 500 tuners per SDV SG service VOD, HSD or VOIP service group Boxes Per Home =1.4 average splitting has become an organic response to Tuners Per Box = 1.3 average address the changing demographics 8 QAMs (256) @ Payload of 37.5 Mbps encompassed within the cable footprint. 10 SDV streams per QAM at 3.75Mbps Diligent monitoring of the various contention 1000HP *0.268 D*1.4STBs/H*1.3 T/B = 520 2 nodes per VOD Service Group Tuners per SDV SG Digital Penetration 45% (Digital 8 QAMs/SG * 10 streams=80 streams/SG Households) Boxes Per Home = 1.8 80 streams/SG / 526 Tuners per SDV SG = Tuners Per Box = 1 (DVRs can only order 0.154 or 15.4% one movie at a time) 4 QAMs (256) @ Payload of 37.5 Mbps Therefore, the 80 unique programs are 10 SDV streams per QAM at 3.75Mbps available to be viewed simultaneous across the 520 digital tuners within the SDV 1000 HP *0.268 D*1.7STBs/H*1T/B = 400 Narrowcast Service Group. Tuners per VOD SG 4 QAMs/SG * 10 streams=40 streams/SG The SDV service is considered a non- blocking service where contention is not 40 streams/SG / 400 Tuners per VOD SG = allowed and must always be available. To 0.099 or about 10% borrow an adage from the telcos, POTs (Plain Old Telephone) now represents Plain Old It is interesting to note that when Television and SDV must provide that same basing VOD math on Digital Homes we have seamless service that the customers expect. a contention rate of 8.9% but when it is SDV differs from StarOver/VOD service, as it normalized to tuner math is around 10% for is a multicast narrowcast service, meaning exactly the same service contention level. So that there is a one-to-many relationship it seems that SDV and VOD should be able so between narrow bandwidth use and user. That share the same physical Narrowcast is to say, within the narrowcast more than one downstream infrastructure even though user can tune into the SDV narrowcast stream running in separate Narrowcast Bandwidth and watch the same program. Going forward, traffic modeling SDV is a unique service as it highly should be calculated using tuner math to dependent on content placed into the SDV normalize the analysis across narrowcast tier. Content is the primary factor in services. determining the utilization of the service and the required bandwidth. For SDV, the number Unifying the Architecture of programs put into the SDV tier will determine the exact amount of bandwidth A couple of questions need to be asked required to support the number of streams. If when architecting a unified narrowcast there are only 80 services in the SDV tier then architecture Voice, Video and Data. there is no possible contention, it is just an expensive broadcast/narrowcast network. Does it make sense to have one SDV becomes much more interesting when it Narrowcast for all services sharing the same is oversubscribed with content. downstream infrastructure physical layer or are there many? If SDV contention is based on tuner math and not Digital Homes, then VOD Can the narrowcast services share the contention should be normalized to that same narrowcast bandwidth? basis. When all services are normalized to Re-calculating VOD Math based on tuners: tuner math, either QAM tuners or DOCISIS 1000 Homes per node tuners, the math and traffic analysis becomes an even more interesting exercise in “And & This will be because video narrowcast Or “ math and understand the peak trending of services under-utilize narrowcast bandwidth services. during the daytime and that excess capacity could be switched over for commercial HSD A single QAM tuner can either be use during the day and then back to residential tuned to a broadcast stream OR to a VOD video narrowcast services in the evening. stream OR to a SDV stream OR turned off. A QAM tuner cannot tune to multiple services at Combined Narrowcast Services the same time. Therefore, it is easily conceived that for services targeted at QAM There are two key technologies that tuners the services can coexist and are missing to truly unify the Narrowcast interoperate quite well within the same shared Services, the Global Session Resource Narrowcast. OR math dictates that is just a Manager (GSRM) and the Business Rules series of tradeoffs. Therefore, when the Engine (BRE). bandwidth for QAM tuner Narrowcast services are shared or pooled together there The Global Session Resource Manager are economies of sharing. (GSRM) is the unifying manager of all the source signal and bandwidth resources. The Current DOCSIS tuners only tune to GSRM negotiates and arbitrates between all one frequency at time. With DOCSIS 3.0 and services and all requests. It is the key channel bonding the DOCSIS tuners can tune bandwidth allocation mechanism; employing wideband frequencies but as they are tuning bandwidth optimization algorithms to ensure discrete frequencies at any one point in time that efficiencies are realized across the they can be considered bound by OR math utilization of bandwidth across all narrowcast when operating in a shared bandwidth pool services. just like they are today. Typically a single 6 MHz narrowcast channel is allocated for HSD The fact that the GSRM will be able to and VOIP phone service across the cable share the narrowcast bandwidth will allow for plant. greater bandwidth usage efficiencies across the combined services and is predicted to Today in the residential market, the require less total bandwidth for the same coexistence complexity arises when looking at blocking factor for any given service. the peaks of QAM tuner services and Efficiency is the key to performance. DOCSIS tuner services. While each service type may be bound by OR math when viewed The Business Rules Engine is the together the two services actually peak at or “uber” Policy Manger for all narrowcast very close to the same time and are thus services. actually bound by AND math. Thus, the total load between a QAM tuner and a DOCSIS It is the tool that determines how to tuner is cumulative. There are not economies “sell” the narrowcast bandwidth for how of scale to share the bandwidth between the much, to whom and prioritizes services and two tuner service types. customers. It “plugs” into the GSRM and is not so much an engineering tool as a business In the near future, when cable has tool. It will allow the cable business to more greatly penetrated into the commercial optimize its service, its services and its market with its HSD offering there will be revenues. some advantages in sharing the bandwidth between video and HSD narrowcast services. Along with the adoption of the GSRM Ōcast Streams Total and BRE a holistic approach must be taken Service 6-MHz Per ŅcastÓ Cumulative into account when architecting the Narrowcast Service Groups Channels Channel Streams Streams design. The physical layer of the Narrowcast 1 Analog 1781 78 78 2 Digital 1 25 12 300 378 cannot just be thought of in an abstract way Same as 3 Ad Zones 4 above 68 272 650 without really identifying with the physical 4 HSD 338 1 Best Effort 338 988 Same as infrastructure of the headend, hubsite, laser, 5 VOIP 200 above QoS 200 1,188 node and customer’s home. This is purely a 6 VOD 170 4 10 6,800 7,988 Same as practical operational model concern. If the 7 StartOver 508 above 10 20,320 28,308 wiring of the various narrowcast services 8 SDV 678 8 10 52,400 82,548 becomes too complicated to manage and the sheer number of service groups, types of The way to interpret the chart above is service groups, QAMs combining and by culminating the growth of each service distribution networks the field personnel will stream load and adding it to the next tier of not be able to support it. Although not service. detailed in this paper, remember to not forget the complexity of the reverse path traffic Graphing this culminating growth across modeling and its physical combining and services creates a graph that looks like: splitting network which is almost equal to the forward path. Broadcast & Narrowcast Growth Case Study – The Exponential Growth 90000 80000 70000

The growth of services dependent on 60000 the video Narrowcast has dynamically 50000 Series1 increased the total number of streams that are 40000 “cast” into the cable plant; either broadcast or 30000 20000 narrowcast. 10000 0 123456789 As cable technology advanced in steps Service Type (from chart above) from 300MHz, 450MHz, 550MHz, 750MHz, 860MHz and now to 1 GHz and at the same time transitioned from an analog broadcast Therefore, for the medium size market network, digital broadcast network, HSD running all services, the cable system went narrowcast, VOD narrowcast, VOIP from broadcasting of a few dozen analog narrowcast and finally to SDV narrowcast streams to managing over 80,000 streams networks the number of channels, programs across their cable footprint within the last 15 and streams offered in cable system has years. grown tremendously. In the case study, the Narrowcasts for For example in a medium size market: SDV and VOD have an equal number of service groups at 675 and can therefore share the same physical downstream wiring, combining and distribution path. However, the number narrowcast service groups for HSD/VOIP has a total of 538 services groups and is not in parity with SDV/VOD and thus cannot share the same physical downstream narrowcast wiring. The questions arise in this real-life example:

To simplify the implementation and operational model, does it make sense to increase the number of HDS service groups by 137 to bring it into parity with SDV and VOD so that all services can share the same downstream narrowcast wiring?

Is the peak demand for HSD going to naturally to follow the peak demand for VOD and SDV?

Is it cost justifiable, practical?

Conclusion – It is One Network

Narrowcast services are pervasive within the cable infrastructure and they form some of the most revenue generating services offered by the cable company. Voice, Video and Data all share the same common cable network but unique Narrowcast Bandwidth.

To fully license this advantage of the Narrowcast Bandwidth, each service should not be looked at discreetly as a unique isolated and separate narrowcast service but in the whole across all service and the cable plant infrastructure.

The emerging and maturing Global Session Resource and Business Rules Engine will unlock the true economies of sharing narrowcast bandwidth within and between HSD and Video services.

The Narrowcast Services will continue to evolve and new yet to be developed and deployed services will guarantee cable’s future success.

OPENNESS AND SECRECY IN SECURITY SYSTEMS: SM POLYCIPHER DOWNLOADABLE CONDITIONAL ACCESS Tom Lookabaugh, PolyCipher Comcast Cable James Fahrny, Abstract and the range and simplicity of implementations. The PolyCipher Downloadable Conditional Access System provides a new The principle security paradigm employed approach to security in cable: a small in DCAS is “defense-in-depth.” In the paper hardware footprint is used to enable both we explain how this contrasts and interacts high security and the flexibility of software with other important concepts in security downloadable security clients for set-top design, including “security by design,” boxes and other cable-ready devices. An “security by obscurity,” Kerckhoffs’ important and often asked question is how principle, the economics of attack and much of this system specification should be defense, effective approaches to security disclosed publicly? Here we explain the system review and qualification, and the role tradeoffs involved in disclosure and motivate of open source, cryptographic primitives, and the ultimate choice: a combination of public modes of standardization for security cryptographic primitives embedded in a systems. private defense-in-depth system. The resulting choices are intended to INTRODUCTION produce a robust security system approach that can achieve both lower costs for the PolyCipher is developing with Cable industry and appropriate security to maintain Television Laboratories a proposed the industry’s unparalleled access to high foundation for downloadable conditional quality content over the coming decades. access (DCAS) for the U.S. cable industry. This system represents a major departure in DOWNLOADABLE CONDITIONAL ACCESS cable security system design, opening up the possibilities of lower cost and increased The PolyCipher Downloadable flexibility in managing access to cable Conditional Access System (DCAS) is an content, while maintaining backwards emerging architecture designed to bring compatibility with the industry’s installed increased power and flexibility to the cable base of security equipment. industry's effort to combat the piracy of its video, audio and other content. In designing a major security system, there are important questions on what kinds Specifically, the PolyCipher DCAS of information are made public and what is architecture is focused on delivering security- kept secret. The choices made affect both the related software clients to basic security of the system and the industry

• Cable-ready mobile and portable compliant cable-ready hosts, including: devices

• Set-top boxes and devices • Other emerging products

• Cable-ready televisions Content security for these devices has

traditionally been handled via hardware, • Home entertainment systems through some combination of set-top devices or the installation of CableCARDTMs. used for encrypting and decrypting the video Unfortunately, this hardware module-driven and media protected by the SM clients. approach requires significant manual effort to upgrade or change security systems at the Software Architecture cable customer level. Furthermore, hardware The PolyCipher DCAS specification defines many key elements: modules must be shipped, inventoried and repaired, all of which drives up operating • Messages between the SM and the DCAS expenses and limits the flexibility of the servers in the headend environment cable operator. • Requirements for SM and DCAS Hosts The PolyCipher DCAS architecture to support DCAS eliminates all this hardware module shuffling because it allows security systems to be • Requirements of the headend server automatically downloaded to compliant devices, using the existing cable • A new key management infrastructure infrastructure. Furthermore, the PolyCipher DCAS specification provides control over a The new key management infrastructure broad range of security-related functions, supports the DCAS architecture by providing including traditional conditional access custom protocols, performance and security systems (CAS: control at the host device of requirements, and by defining the necessary access to content), authorized service domain levels of interoperability, accountability and (ASD: control of other local devices all security. secured by the cable operator’s security system), and digital rights management OPENNESS AND SECRECY IN DCAS (DRM: bridging to other security systems on other local devices). In simple terms, a security system can be thought of as an algorithm and a key. The Hardware Architecture algorithm explains what happens to accomplish the functions desired if the key is The PolyCipher DCAS hardware known. The key is a piece of data that is architecture includes a Secure Micro (SM) kept secret except to those who need to use it and a Transport Processor (TP). The SM is a to operate the system. hardened and limited-capability microprocessor that primarily enables the decryption of multiple video streams, under While there is no debate about the direction of the installed CAS client. It does importance of keeping the key secret, there is this by providing the necessary key an ongoing debate on when to keep the management services for the TP and enabling algorithm secret. a secure bootstrap of the software system [1]. Making the algorithm public allows for a The download of clients (CAS, ASD or broad community with diverse tools and DRM) to the SM is securely managed in the perspectives to carefully evaluate it. Many network operator's headend via the an assistant professor or graduate student can interaction of the SM software and a DCAS win glory (and maybe tenure) by finding a authentication proxy. The TP is primarily critical flaw in a proposed algorithm. The notion that a security system should only rely on the secrecy of the key for its security goes all kept secret, which are sequentially by the name Kerckhoffs’ principle [2, 3]. A deployed when and if previous deployed frequent complement to Kerckhoffs’ counter measures are breached. The idea is principle is the notion that reliance on the that with each deployment, there is an principle can only be reasonably assured by extension of the period of low gain for submitting a security system to broad public pirates and low loss for the operator. scrutiny. So, what should reasonably decide which Many important security primitives kind or which parts of a security system (algorithms focused on a narrow security algorithm should be public versus secret? function) are publicly vetted in exactly this way, examples include AES, RSA, and a A useful answer is: when the system is variety of others that are used commonly in simple enough that the cost to evaluators to security systems of any scale. Some, like test it is more than offset by their expected DES, were originally created in secret but gains (in money or, more typically, fame and later subjected to intense public scrutiny. reputation), the public scrutiny that is frequently equated with Kerckhoffs’ More generally, software systems such as principle will be valuable. Conversely, when operating systems and browsers have been the system is complex enough and of limited the subject of a debate on the relative merits enough interest that public scrutiny will only for security of open and closed source. result in incomplete vetting, public disclosure These systems can be quite complex but are may backfire, since an attacker needs find also of intrinsic interest to a large community only one flaw in a public algorithm to breach of developers. There does not seem to be a it (while the evaluators have the much harder definitive answer as to whether such systems task of attempting to test and eliminate all (complex but of broad interest) are more possible flaws). As Schneier points out, secure in open or closed source form, there are limits to the amount of gratis work although a number of authors give the edge one can expect of the security community: to recommending open source for security [4, “Security researchers are fickle and busy 5, 6, 7]. people. They do not have the time, nor the inclination, to examine every piece of source However, many large security systems do code that is published.” [9, pp. 343-346]. Or, not fully subscribe to public scrutiny but in Anderson’s words: “Arguments against choose, instead, to keep substantial parts of open source center on the fact that once the security system algorithms private. A software becomes large and complex, there relevant alternative paradigm, especially may be few or no capable motivated people common in pay-TV systems, is called studying it, hence major vulnerabilities may “defense in depth” [8]. This approach views take years to be discovered….the important security as an economic competition between questions are how much effort was expended the security operator and his opponents by capable people in checking and testing the (pirates and hackers); the goal is not perfect code – and whether they tell you everything security – which is deemed to be they find” [8, pp. 296-207]. unachievable – but rather a situation in which the economic gain to the pirates and The Data Encryption Standard (DES) economic losses of the security operator are encryption algorithm, for example, can be both sufficiently low to maintain the viability described in about 100 lines of source code. of the operator’s business model. Defense in It is straightforward for academics, depth envisions a series of counter measures, hobbyists, and professionals to understand and analyze it in detail from many different with a dependence on algorithms’ secrecy for perspectives. The same is true of many their success? Ultimately, the analysis there cryptographic primitives. is economically motivated – as in this paper: if defenders profit more from exposure than The protocols, procedures, and algorithms attackers, then disclosure is valuable; if not, involved in a full scale conditional access then not. Note again, though that Kerchoffs’ system sit near the other end of a continuum. principle is always valuable in design. Describing these fully could easily run to thousands of pages of documentation. It is The implications for open standardization so expensive to fully comprehend these that of security systems follow. Security the amount of evaluation that can be primitives certainly benefit from the expected (other from those explicitly paid to evaluation possible in an open standards do so) is quite limited. And the intrinsic setting (although this is not the only way to motivation for the security community to obtain public scrutiny – for example, an protect a particular instance of a conditional alternative is to publish a patented algorithm access system could reasonably be expected and provide a prize to those who breach it). to be much less than that for a widely used Large scale security systems benefit if they application like an operating system or proportionally scale in their interest to the browser. Moreover, systems this complex are security community – so experts find it worth simply never bug free; the combinatorics of their while to provide free scrutiny. Less analysis and testing make this infeasible. interesting systems, such as a particular The result is that such systems are rarely if conditional access system implementation, ever made public. This does not mean there do not benefit from open standardization – in is an intention that the secrecy of the the sense that security is weakened rather algorithm is its sole defense; indeed than improved in the particular and deciding Kerckhoffs’ principle is as much an objective context of system economics. This leaves here as it is in a cryptographic primitive. But open the exact boundary between a security both the reality of large scale system creation primitive and a large scale system, but the and system test and the particular economic principle is clear. incentives of system creators, pirates, and potential reviewers mean that striving for the PolyCipher Approach goal of Kerckhoffs’ principle is supplemented by the use of defense in depth The PolyCipher approach attempts to find to manage the economics that are the the best combination of open, public fundamental driver in protecting a algorithms and closed, defense-in-depth commercial conditional access. system design. Cryptographic primitives used in the system design are drawn from Swire has developed a thoughtful analysis those that are widely used and well of the economic, legal, and regulatory established, such as the DES and AES implications of tradeoffs in security system encryption algorithms. disclosure in a pair of papers [10, 11]. There he provides a useful comparison with The overall system is not public, though, military cryptography (remember, and is divided into tiers of system design and Kerckhoffs was in fact addressing military documentation that are increasingly access uses): why is it that militaries consistently restricted. Although the overall system is too find it valuable to keep cryptographic complex to expect that any substantial gratis algorithms secret, even while adhering to vetting would be applied if the system were Kerckhoffs’ idea that they shouldn’t design made public, paid vetting by experts is extensively applied. Additionally, [5] Crispin Cowan, “Software Security for stakeholders who will depend on the security Open-Source Systems,” IEEE Security and of the system and who are themselves expert Privacy, January/February 2003, pp. 39-46. in this type of security are invited to audit the system design. Motivations for stakeholders [6] Rebecca T. Mercuri, “Trusting in vary and in many cases they can be expected Transparency,” Communications of the to be quite critical of the system. The ACM, May 2005, pp. 15-19. resulting review environment is contentious but thorough. [7] Jaap-Henk Hoepman and Bart Jacobs, “Increased Security Through Open Source,” CONCLUSION Communications of the ACM, January 2007, pp. 79-83. The PolyCipher DCAS system offers a new strategy for providing hardware rooted downloadable security for the cable industry. The question of how much of the system to publicly disclose is an important one and [8] Ross Anderson. Security Engineering: A requires careful thought. But after Guide to Building Dependable Distributed consideration, the complexity of the system Systems, New York: John Wiley & Sons, Inc. and narrowness of application suggests a 2001. hybrid approach: the use of well vetted publicly known cryptographic primitives [9] Bruce Schneier, Secrets and Lies: Digital embedded in a “defense in depth’ system Security in a Networked World. New York: subject to extensive but private review. John Wiley and Sons, Inc. 2000.

REFERENCES [10] Peter Swire, “A Model for When Disclosure Helps Security: What is Different [1] William A. Arbaugh, David J. Farber, about Computer and Network Security,” Jonathan M. Smith, “A Secure and Reliable Journal of Telecommunications and High Bootstrap Architecutre,” in Proceedings of Technology Law, vol.2, 2004. the IEEE Symposium on Security and Privacy, May 4-7, 1997. [11] Peter Swire, “A Theory of Disclosure for Security and Competitive Reasons: Open [2] August Kerckhoffs, “La cryptographie Source, Proprietary Software, and militaire,” Journal des sciences militaries, Government Agencies,” Houston Law vol. IX, pp. 5-38, Jan. 1883, pp. 161-181, Review, vol. 42, No. 5, January 2006. Fev. 1883. AUTHORS [3] Bruce Schneier, “Secrecy, Security, and Obscurity,” Crypto-Gram Newsletter, May Tom Lookabaugh is with PolyCipher, 999 15, 2002, available at 18th St., Su. 1925, Denver, CO 80202, and by http://www.schneier.com/crypto-gram.html. email at [email protected].

[4] Brian Whitten, Carl Landwehr, and James Fahrny is with Comcast Cable, 1500 Michael Caloyannides, “Does Open Source Market St., Philadelphia, PA 19102 and with Improve System Security,” IEEE Software, PolyCipher, 999 18th St., Su. 1925, Denver, September/October 2001, pp. 57-61. CO 80202. He can be reached by email at [email protected]. PRACTICAL VIDEO OVER DOCSIS® IMPLEMENTATIONS WITHOUT FORKLIFTS Doug Jones, Chief Architect Roger Slyk, Director, Product Marketing BigBand Networks

Abstract • Maximizing the seamlessness of the

customer experience; This paper describes a scalable IPTV (video over IP over DOCSIS) solution for • Placing the intelligence in the network, providing video content to a wide array of and minimize the intelligence required consumer devices that span TVs and PCs, and in set-top boxes, to protect CPE other device capable of receiving and (customer premise equipment) processing video streams over IP. The investment and maximize flexibility. solution is extensible to other last-mile edge The authors advocate that, while there networks including 3GPP, Wi-Fi, and WiMAX may be many paths to an IPTV reality, the wireless networks, as well as going “over the best evolution plan utilizes M-CMTS top” on IP access networks managed by third- (Modular CMTS) designs that leverage party providers. It defines and describes a existing DOCSIS 2.0 infrastructures while distribution network that sources live accelerating towards DOCSIS 3.0 broadband broadcast and stored content from VOD speeds and capabilities. The authors describe (Video on Demand), PPV (Pay Per View), and an evolutionary strategy based on an M- Internet sources, with elements of centralized CMTS approach that can drive the and distributed processing and management. availability of downstream DOCSIS 3.0 The proposed distribution network utilizes channel bonding, enable more flexible existing network infrastructures and CMTS allocations of upstream to downstream traffic (cable modem termination system) platforms flows, and provide additional performance as the basis for the introduction of advanced benefits. Additionally, other parts of the services delivery, thus avoiding significant solution are already specified by the cable equipment change out, capital expense and industry, including PacketCable™ Multimedia operational disruption. and Embedded DOCSIS.

The authors describe key business and INTRODUCTION technology requirements that are intended to be a reference when making decisions about Consumers are rapidly demanding an the IPTV system design in order to maintain experience that blurs the line between “lean- the business viability of the solution. These back” consumption of TV-based include: entertainment and “lean-forward” multimedia • Ensuring the solution is agnostic of the activities on their personal computers. It is last mile network; not a single killer application, nor an intentional move towards new technology that • Enabling scalable, flexible and cost is driving this rapid change in consumer effective systems for wide scale behavior, but a merging of technology, deployment; content and consumer acceptance. The ability to watch similar content in a home theater and on a portable device is increasingly being traditional STBs (set-top boxes) over more demanded; the capability to access a broad than just the HFC (Hybrid Fiber-Coax) array of content from many service providers network. Cable services are evolving to is rapidly becoming expected as well. Cable include to new services that will be delivered operators must plan for the flexible delivery to a new generation of devices such as mobile of any content from any source to any phones, wireless devices and personal consumer device in order to remain computers. The network infrastructure will competitive in the long term. An open need to be capable of obtaining content from architecture based on IP (Internet Protocol) is broad range of sources including broadcast the preferable long-term solution. programming, VOD, PPV and emerging “new media” outlets such as YouTube, MetaCafe and Ziddio. Cable operators need the capability to deliver their services to more than just

Figure 1: Delivery of video content over DOCSIS infrastructures is increasingly popular

But there is a great incentive to have as could include 3GPP (3rd Generation Party many of the new services as possible Partnership) wireless networks, or any kind of delivered to the existing base of MPEG-2 high-speed IP access network which the STBs as well as to a new generation of CPE operator does not even own. with advanced encoding technology and high- speed interactive communications channels To incent customers, this solution will based on DOCSIS and DSG (DOCSIS Set- offer broader ranges of content, time-shifting Top Gateway). and place-shifting and greater opportunities for personalization The proposed network architecture to deliver these services is based on a core IPTV SERVICES network of video and communications services which provide IP-in, IP-out transport An IP service is a cable service delivered and processing and any number of edge to an end user over an edge network capable networks which provide IP transport to the of transporting IP. This makes no statement customer. The delivery of IPTV services over to the origination of the content; that is, it the existing HFC edge network will be based could be a traditional video service that an on high capacity, cost-efficient M-CMTS operator offers today, for example linear components, as well as low cost, scaleable television service, or it could be a video clip MPEG-2 edge QAMs. Other edge networks from a so-called new media outlet such as a

website or some other Internet-based source. begin laying the groundwork to deliver all These are both video services and they can be cable services, including video, over the HFC packaged into a service based on an operator edge network using IP. business plan or objective, but either can be delivered to the consumer using IP. NEXT GENERATION IPTV INFRASTRUCTURES: A VISION IPTV services generally exhibit several characteristics including: The figure below provides a reference model for Internet TV that includes a core • content personalization through time- network and one or more last mile edge shifting, place-shifting and addressable networks which has the goal of delivering advertising; cable services to a variety of end-user devices over a variety of edge networks. This • expansion of programming choices by network can deliver linear, switched and On broadening the universe of content to Demand services. The diagram does not include new media services available imply a one-to-one relationship between over an Internet connection; components. • ubiquity of services due to the prevalence of networks which support IP transport.

Today, HSD and Voice are offered over the HFC network as IP services. The only cable service not offered over IP on the HFC is video although video services are distributed around cable networks (for example backbones and regional area networks) using IP, but delivered to the end Figure 2: Internet TV reference network user over the HFC edge network using MPEG-2 transport. The Ingest, Distribution and Following the trends, video will also go Personalization functions are considered part over IP on the HFC and one reason to do this of a core network which can provide a variety is to prepare to offer cable services on other of content and services to a variety of IP- edge networks which could include wireless capable edge networks. The system uses IP networks or other wired networks, with the networking and transport as both the method common denominator that they support IP for propagating content and as the method for transport of cable services. It is also expected staging content along the various steps of the that services will be delivered over networks system. which are not necessarily owned by the cable The paper first describes the HFC last operators themselves (for example, these mile edge network for supporting IPTV services would be provided “over the top” on services, including M-CMTS and DOCSIS a network owned by third-parties). 3.0, to illustrate how these solutions work Getting back to HFC, with Modular together to support high capacity, high CMTS, DOCSIS 3.0 and other specifications bandwidth IP services to consumers. This such as PacketCable Multimedia and section also describes how IPTV services Embedded DOCSIS, the toolset is available to could be delivered to existing MPEG-2 STBs. The core network provides functions which customer premise equipment. 3GPP wireless are similar to existing cable networks. Later technology delivers broadband-like data the paper posits how these core network speeds to mobile devices while allowing functions are used to deliver cable services in consumers to send and receive services an all-IP environment. including data, voice, digital images, web pages, photographs and video three times faster than possible with a traditional EDGE NETWORKS GSM/GPRS network.

As alluded to earlier, various edge networks are possible and several are shown in the diagrams that follow.

Figure 5: IEEE 802.x edge network

IEEE 802.x refers to a family of specifications developed by the Institute of

Electrical and Electronics Engineers for wireless LAN technology. IEEE 802.11 (also Figure 3: DOCSIS edge network known as Wi-Fi) specifies an over-the-air interface between a wireless client and a base The figure above shows the DOCSIS edge station which supports data rates up to 54 network at a high level of detail, and includes Mbps per AP (Access Point) over a distance the PCMM (PacketCable Multimedia) policy of up to 100 meters. IEEE 802.11 generally server and a session management function. requires line of sight access and is suitable for The DOCSIS Edge network will be described hotspot access in areas of a city. in more detail in a later section. IEEE 802.16 (also know as WiMax) refers to a family of specifications developed by the IEEE for Broadband wireless access that provides secure, full-duplex, fixed wireless MAN (Metropolitan Area Network) service. Throughput can reach 75 Mbps and does not require line-of-sight to operate and reach can extend from one mile at full speed to thirty miles at reduced throughput. Figure 4: 3 GPP wireless edge network IEEE 802.20 (also know as Mobile-Fi) is The 3GPP wireless edge supports third similar to WiMax but includes optimization generation mobile network services which for roaming and hand-off at speeds up to 150 include the ability to deliver higher data rates miles per hour. and advanced media features. The 3GPP DOCSIS 3.0 / M-CMTS Edge Network wireless edge supports native IP services to

The advantages provided by DOCSIS 3.0 enable an operator to provide IP video services over their HFC networks. DOCSIS in general provides IP transport over HFC networks and DOCSIS 3.0 in conjunction with PCMM provides the functions needed to deliver high-speed video services over IP over an HFC network. Figure 6: Conceptual view of M-CMTS architecture DOCSIS 1.1 provided the necessary QoS (Quality of Service) for IP services and Original CMTS implementations placed DOCSIS 2.0 provided higher capacity return the CMTS core and QAM functions in an path channels. DOCSIS 3.0 provides channel integrated assembly, fed with a common bonding services which dramatically increase internal clock. The M-CMTS architecture both upstream and downstream speeds and creates interfaces to separate the downstream capacities as well as the IP Multicast services edge QAM from the M-CMTS core. It is this needed to deliver video. The DOCSIS 3.0 separation which allows massive, scaleable specifications also describe how existing downstream capacity through the use of DOCSIS 2.0 and DOCSIS 1.x cable modems universal edge QAMs instead of downstream operate on DOCSIS 3.0 upstream and QAMs integrated into the CMTS as is the downstream channels. The combination of case today. DOCSIS 3.0 and M-CMTS provide both the technical and economic solutions to providing Since M-CMTS enables the separation of video services over DOCSIS. these components, the DOCSIS Timing Generator was created to supply the accurate The M-CMTS specification describes an and robust transport of a master clock signal economical method for implementing between the M-CMTS components. The downstream channel bonding which is a key DOCSIS Timing protocol is structured to enabler for offering entertainment quality support all SCDMA and TDMA timing video over DOCSIS. requirements. In the M-CMTS architecture shown An added benefit of an M-CMTS below, a device referred to as the M-CMTS architecture is its ability to support flexible Core contains the DOCSIS MAC (Media allocations of upstream and downstream Access Control) functions which include all channels to meet changing traffic pattern DOCSIS signaling functions, downstream demands. That is, M-CMTS can offer more bandwidth scheduling, and DOCSIS framing. capacity in the downstream direction for A UEQ (Universal Edge QAM) contains video services during the evening hours, but mainly physical layer related circuitry, such symmetric upstream and downstream capacity as QAM modulators, and packet tunneling during business hours when commercial data logic to connect to the M-CMTS Core. services are at peak usage. ACHIEVING SCALEABLE DOWNSTREAM DOCSIS CAPACITY

Delivering traditional cable video services over DOCSIS requires a lot of downstream capacity. M-CMTS allows downstream speeds by factors ranging from 2 to 20, to capacity to be added to the system by adding support services ranging up to HDTV (High scaleable modular UEQs to the M-CMTS Definition Television). core.

The original integrated CMTS implementations included cards that had both upstream and downstream capabilities. These implementations were suitable for the relatively low bit rate services of high speed Internet and voice. But video services changed the requirements of the system by requiring significantly more downstream capacity than before. Integrated CMTS implementations cannot provide massive economic downstream capacity because each time a card is added, the upstream capacity of those cards is left stranded. Even a downstream-only card is not optimal because Figure 7: Downstream M-CMTS traffic flows of the opportunity cost associated with sparing the additional downstream-only card A key element of downstream channel in the CMTS chassis to provide redundancy bonding as described in DOCSIS 3.0 is the for services. ability of existing DOCSIS 1.x and DOCSIS 2.0 CMs to operate transparently on channels The M-CMTS specification was designed which are bonded together for DOCSIS 3.0 to address these two issues and bring a third operation. solution to bear by supporting the use of dense, scaleable UEQs to provide the Additionally, DOCSIS 3.0 allows the M- downstream capacity while offering the CMTS to assign individual downstream opportunity to share the UEQs with other services to particular downstream channels. services such as traditional digital video, For example, an M-CMTS may assign a switched video and video On Demand. video-over-IP service flow to a downstream channel with deeper interleaving for higher DOCSIS 3.0 describes channel bonding reliability, while also assigning a voice services which supports the concurrent service destined for the same modem to a scheduling of DOCSIS data over multiple different downstream channel with shallower channels. interleaving for low latency. As shown in the following figure for downstream channel bonding, the CMTS DELIVERING IP VIDEO OVER HFC distributes individual IP packets over multiple NETWORKS QAM channels in such a way that permits the CM (cable modem) to resequence out-of- An M-CMTS architecture can be order packets. By bonding together multiple combined with DOCSIS 3.0 to economically downstream channels, the aggregate deliver cable video services over a DOCSIS downstream speeds to a single CM can be 100 network, including linear, switched digital Mbps, or more, and have potential to increase and On Demand services. broadband subscribers' downstream access

Session Delivering Linear and Switched Digital IPTV Management Services PCMM Policy Server The DOCSIS 3.0 specifications describe a 3a 2 1 method for delivering linear IPTV services; M-CMTS Universal Core Network CM Client however, it is the addition of PCMM which core Edge QAM allows service guarantees to be offered 3b through policy-based QoS management on the HFC network. Figure 8: Synergy between PCMM and IPTV Policy-based QoS management ensures the bandwidth needed for the video service is To change to a linear channel, the IPTV allocated and managed on the HFC network. client sends an IGMP Join (1) which is Without this management, it would be received by the M-CMTS core. If that possible to put more services on the HFC than channel is not already on a downstream can be carried by the available bit rate, that is, accessible by that cable modem, but is already video services could interfere with data or flowing into the M-CMTS core, upon receipt voice services, and visa versa. It is the of the IGMP Join the M-CMTS will initiate a PCMM function which allows the operator to query (2) to the PCMM policy server to manage the HFC resources and make allocate capacity on the appropriate intelligent business decisions in times of downstream for that service, then configure resource contention as to which services the CM for the IP multicast flow carrying that should be allowed on the HFC network. service (3b) and then forward the service to that cable modem. There are a couple of models being investigated which use PCMM for delivering If that channel is already on a downstream linear services and the intent is to provide the accessible by that cable modem, the M-CMTS best customer experience by minimizing core will configure the CM for the IP channel change times while providing the multicast flow which is already carrying that operator with the necessary control points to service (3b) on the appropriate downstream manage the service. and the client will be able to view the content. This is an example of the inherent video Both methods exhibit a key feature of switching capability of DOCSIS 3.0. Once a DOCSIS 3.0 which is the ability to natively linear service is on a downstream channel, provide switched digital video as a by-product any client which can access that downstream of the IPTV implementation using IP channel can access that service; hence, only Multicast. one copy of the service is needed on that downstream even if there are multiple viewers The diagram below shows the first method of that service. To complete the transaction, which follows a traditional IPTV model the M-CMTS core will signal the PCMM adapted to use PCMM where the IPTV client policy server of the channel change request. sends an IGMP Join to the network to “change channels” to a broadcast service. If that channel is not already on a downstream accessible by that cable modem, and is also not yet flowing into the M-CMTS core, upon receipt of the IGMP Join the M- CMTS will first complete the query (2) to the PCMM policy server to allocate capacity on view the content. This also is an example of the downstream for that service and then issue the inherent video switching capability of an IGMP Join to the core network (3a) to get DOCSIS 3.0. Once a linear service is on a the service to flow to the M-CMTS core, then downstream channel, any client which can configure the CM for the IP multicast flow access that downstream channel can access carrying that service (3b) and then forward that service; hence, only one copy of the the service to that cable modem. service is needed on that downstream. Notice that there is no explicit interaction If that channel is not already on a with the session management function. Since downstream accessible by that cable modem, all channel change requests are signaled to the but is already flowing into the M-CMTS core, PCMM policy server, session information can the M-CMTS will configure the CM for the IP be queried from there. This is in contrast to multicast flow carrying that service (4b) and the second way of implementing linear IPTV then forward the service to the downstreams services with PCMM which explicitly uses which services that cable modem. the session management function. If that channel is not already flowing into The diagram below shows a linear IPTV the M-CMTS core, the M-CMTS will first model with a session-based controller adapted issue an IGMP Join to the core network (4a) to use PCMM. Interaction with session to get the service to flow to the M-CMTS management is very similar to switched core, then configure the CM for the IP broadcast. In this model, when a channel multicast flow carrying that service (4b) and change occurs the IPTV client issues a session then forward the service to that cable modem. request to the session management function. Note that in this case there is always an interaction with a session server, as well as interaction with the PCMM policy server.

Delivering On Demand IPTV Services

The DOCSIS 3.0 specifications support the delivery of On Demand IPTV services as unicast streams. As with linear services, the addition of PCMM allows service guarantees to be offered through policy QoS management Figure 9: Linear TV model with session-based on the HFC network. control The model for On Demand IPTV services Upon receipt of the channel change is very similar to existing On Demand request, the session management function services, except rather than send the stream interacts with the PCMM policy server to through a video QAM it is sent through a request bandwidth on the M-CMTS core for DOCSIS QAM. the service. If that channel is already on a downstream accessible by that cable modem, The diagram below shows an On Demand the M-CMTS core will configure the CM for IPTV model with a session-based controller the IP multicast flow which is already adapted to use PCMM. In this model, the carrying that service (4b) on the appropriate IPTV client issues a session request to the downstream and the client will be able to

session manager when requesting On Demand The following figure shows a solution content. utilizing a stand-alone DOCSIS CM connected to the Client over a home network.

Figure 10: On Demand IPTV with session- Figure 12: Stand-alone cable modem and based control client

The session management then instantiates In this case, the home network supports QoS with the M-CMTS core through the video streaming in order not to have a PCMM policy server (2, 3). If the bandwidth negative effect on the service. needed to deliver that service is available, the With all DOCSIS CM solutions, an M- session manager will instruct the core network CMTS solution with downstream channel to begin streaming the service to the client. bonding would provide the most scaleable and Considerations for a Cable IPTV STB economic solution. Downstream channel bonding provides additional capacity which An IPTV STB needs a connection to a will benefit video services. Additionally, DOCSIS cable modem. It also needs to be DOCSIS/DSG is used for the out-of-band able to decode video from the CM to the communication path thereby obviating need decoder. for legacy return path components in the STB. The following figure shows an example of The following figure shows an Off-Net an eDOCSIS (Embedded DOCSIS) STB CPE solution where the last mile access IP solution where the DOCSIS CM follows the network is provided by a third party. In this eDOCSIS specifications and the Client is case, a third party manages the network and considered an Embedded DOCSIS Embedded the relationship with the NID (Network eSAFE (Service Application Functional Interface Device). The cable operator is only Entity). a provider of content services to the client.

eCM Client

HFC Network Internal Connection eDOCSIS connection Figure 13: Off-Net CPE in over-the-top application Figure 11: Embedded DOCSIS STB solution The client will need the ability to decode The above figure indicates how content video in the format that is sourced from the available in the core network can be cable operator. IPTV content can be encoded distributed to an existing video QAM just as and delivered in several formats. The easily as a DOCSIS QAM through an M- traditional format for cable digital video is CMTS. MPEG-2 transport streams over UDP/IP, specifically packing seven 188-Byte MPEG-2 The session management function will cells in one IP packet. This is the same know the capabilities of the entity requesting format used to deliver On Demand or digital the service, whether it is an IP client or an simulcast content over Gigabit Ethernet to a existing STB application. When the entity QAM. This same format can be used by an places the request, the core network will M-CMTS to deliver digital content to a stream the content over IP to the HFC DOCSIS QAM for consumption by the client. network. If the content is destined to an IP client then the system will route the content An alternative method to deliver IPTV over the M-CMTS QAM. If the content is may be supported by “new media” sourced destined for an existing STB, the system will from Internet content providers. In this case, route the content over IP to a video QAM just the content is coded as elementary streams as happens today for digital simulcast and On over RTP/UDP/IP with no MPEG-2 framing. Demand. The video QAM will take the In this case, the timing for the content is content over IP as input and then strip off the recovered from the RTP layer instead of from IP and deliver the stream as native MPEG just the MPEG-2 transport stream layer. This as it does today to digital STBs. method is how content is delivered to clients such as Windows Media and Real Player and Existing digital STBs can only render there are standards available for delivering content which is encoded as MPEG-2 content using MPEG-2, H.264 and AC-3 transport streams. Some IPTV content can be codecs as well; hence, this may be an encoded as elementary streams over IP, with emerging trend for IPTV content. no MPEG-2 framing. For delivery of codec elementary streams to an existing STB, there will need to be a transcoding step to change Delivering IPTV Services to Existing STBs the codec elementary streams over IP to an MPEG-2 transport stream over IP. This The IPTV architecture could be function is currently defined to happen as part considered incomplete if it did not address of the personalization function. delivering IPTV services to legacy STBs. With this transcoding function, it will be possible to deliver IPTV content, regardless of the encoding and transport, to both IPTV devices and to existing STBs. CORE NETWORK

The functions of the core network show many similarities to existing cable networks, including linear and On Demand services. The differences are in the areas of making the Figure 14: Video delivery to Edge QAMs existing and emerging services available to a wider array of end devices. In all cases it is assumed the transport of content to these end

devices uses IP, but these devices will have generates content in the proper format/profile characteristics which vary from those of the and with the proper metadata for the media traditional STB. devices supported in the network. The core network can be characterized by Minimally the transcoding system must three functions including Ingest, Distribution support the transcoding of MPEG-2 video and and Personalization. audio to H.264 and MP3 or AAC standards, respectively. In addition, the transcoding Ingest Functions system may support conversion from H.264 to MPEG-2 for providing internet-sourced The Ingest functions of the core network content to existing STBs. Both SDTV and are shown in the following figure. HDTV formats should be supported. For use with mobile networks, the transcoding system should support the media formats that are described in the PacketCable Release 2.0 Codec and Media specification, available through CableLabs. This specification requires the H.263 codec as well as recommends various profiles of H.264 as optional.

Figure 15: Ingest functionality in network Depending on operator plans, additional model codecs could be considered including the transcoding of MPEG-2 video into the The Ingest functions are designed to following formats: acquire content from a variety of sources including linear broadcast networks, On • Flash Demand storage and content from sources that might otherwise be associated with the • Windows Media Internet. • Real A key functions of the Ingest network is asset acquisition which includes the ability to • Quicktime connect and retrieve real-time and non-real- time content feeds including the reception and distribution of content from multiple Though these formats are available with independent sources to multiple distribution CDN systems, there are many system servers and applications. implication to supporting these formats which need to be investigated. The asset acquisition system streamlines and automates the process of receiving, Video content carried over IP can take one staging, storing and propagating content of two basic forms. The first encapsulates an including creating index files (for example MPEG-2 transport stream over IP as is used trick mode files, for On Demand content.) today to transport video streams over backbone and regional area networks. The Another important function of the Ingest second encapsulates codec elementary network is transcoding. The transcoding streams directly into RTP over IP with no system processes MPEG-2 content and MPEG-2 framing. These two methods are fundamentally different and are not and provides intelligent up-sell opportunities necessarily interchangeable at end devices by interfacing with the billing system. The such as STBs and mobile video players. The Service Aggregation system makes it easy to core network should include the capability to create new services, push personalized offers, transcode from one to the other while either and recommend add-on services and up-sell. maintaining the same codecs or when switching between codecs. The Service Aggregation function interacts with the Ingest system and obtains The next function includes the insertion of information necessary for the On Demand metadata which should comply with the application from other components, such as CableLabs metadata specification, which may the asset metadata from the Asset need an update to support the new types of Management System. This function of content which can be ingested including Service Aggregation interacts with the Host to obtaining content and metadata from various present navigation menus and related internal and external (for example the application features to the On Demand Client Internet) sources. This function should and exchanges messages with the On Demand support both automatic and manual insertion Client to enable the navigation functions, and include the capability to translate from including parental controls. proprietary metadata formats to the format specified in the CableLabs metadata The distribution technology will be specification. chosen to provide a dynamic system for managing content in the network. Finally, an encryption function should be present to protect content on the core network The most common method by which files and an archive function could be present to are transferred on an IP network is the client- maintain copies of content. server model where a central server sends the entire file to each client that requests it. In this model the clients only speak to the server, Distribution Functions and not to each other. The main advantages of this method are that it's simple to set up, The distribution functions are shown in and the files are usually always available the following figure. since the servers tend to be dedicated to the task of serving, and are always on and connected to the Internet. However, this model has a significant problem with files that are large or very popular, or both. Namely, it Figure 16: Service aggregation and takes a great deal of bandwidth and server distribution resources to distribute such a file since the server must transmit the entire file to each Service Aggregation defines the set of client. The concept of mirrors partially service packages and bundles that are addresses this shortcoming by distributing the ultimately sold to the subscriber. Packages load across multiple servers but it requires a can be composed of real-time transcoded lot of coordination and effort to set up an programs, multicast time shifted programs, efficient network of mirrors and it's usually unicast On Demand programs or a only feasible for the busiest of sites. combination of all of the above. On the newer end of the spectrum are The Service Aggregation system Torrent-based technologies which work by understands the current promotional offers taking a large file and breaking it into pieces

which are spread out over a network of The architecture allows service and servers. When the file is downloaded, it business logic to be incorporated into the travels over the network in pieces from resource selection process. For example, it multiple sources which takes less time than may be more important to fulfill requests for downloading the file from a single source. sessions associated with a pay per view movie The issue of scale with popular downloads is service than for sessions associated with a somewhat mitigated because there's a greater free On Demand Service. The architecture chance that a popular file will be offered by a should be able to translate service and number of servers. Overall, distributing the business logic into a set of rules to be download in this fashion reduces the incorporated in the resource selection process. aggregate load on the servers and makes for a more efficient network. The architecture should make it possible to incorporate new classes of resources along with new resource managers without having Personalization Functions to make major modifications to other The personalization functions are shown components of the system. The architecture in the following figure. should allow the ability to modify or add resources to an existing session. The architecture and resulting protocols should enable the components to be implemented using simple state machines. The architecture should also minimize the amount of state that needs to be maintained across components. Figure 17: Personalization functions

Session Management Addressable Ad Insertion

The Session Manager should support the Addressable advertising provides the ability for components to scale to large capability of addressing specific ads to numbers of active sessions and many session specific customers. An AA (addressable setups / teardowns per second. The Session advertising) architecture is designed to make Manager implements the service and business near real-time decisions on addressability that logic, including policy decisions, associated can differentiate between individual STBs and with Internet TV. The number of session may different users watching that STB. For grow very large and the architecture needs to example, if a home has just one STB the support optimized resource allocation when system should be able to discern if it is a many potential resources for a particular parent or a child using that STB and address session exist. the appropriate ads. For example, the system could use near real-time analysis of remote The architecture should be designed to control key presses to determine if the user is result in minimal session setup latency by watching sports, dramas or cartoons and make minimizing the number of messages used to the appropriate near real-time decisions on set up sessions and allowing as much which ads to insert. The system can also parallelism as possible between the make use of outside information such as components involved in setting up the session. billing profile, demographic and economic data. Addressable advertising is available for all models for delivered content through the services provided over the network. secure distribution of rights to trusted players. The DRM (Digital Right Management) Instant Channel Change technology consists of two components including metadata associated with a piece of Instant Channel Change enables content which describes the business rules channel changing times that are less than associated with that content. These rules traditional cable network tuning of digital generally indicate how the content can be channels. ICC (Instant Channel Change) consumed, by whom, and for how long. effectively eliminates the delay associated Services are built using the metadata. with tuning channels in a digital cable system Additionally DRM includes cryptographic by exploiting the large video buffers in the algorithms used to protect the metadata and STB available with an IPTV system. the content from unauthorized use.

Trans-Coding & Trans-Rating The DRM secures both the content and the business rules associated with that content. The trans-coding function is a final The devices enforcing the DRM must be opportunity to modify the codecs used to trusted, e.g., STBs, PCs, portable devices, etc. deliver the service. and must be manufactured in such a way that the DRM is not easily defeated. The trans-rating function is available to “down rate” the service from the bit rate used The DRM metadata is used to describe for transport on the network to a bit rate that business models and should be flexible is suitable for delivery to the client. For enough to allow an operator to implement a example, a 1.5 Mbps H.264 encoded service wide array of services. For instance, DRM could be trans-rated to 300 kbps for delivery metadata could require a piece of content to over an edge network which cannot support be deleted after 5 plays could disallow trick- higher data rates. mode playback during a certain window of time (for example a theatrical release). Encryption DRM could be designed as an extension of the operators existing conditional access The encryption function is used to make systems, and this should include multiroom the service unreadable by users without capabilities, i.e., sharing content over a home knowledge of the keys to decrypt the service. network and possibly even over large The encryption may be specific to the geographic distances. client and possibilities include Windows Media, Real Helix or any other scheme used IP Streaming by the cable operator. The IP Streaming function will support outputting content in either IP multicast or IP unicast streams. The IP Streaming function must support IPv4 and must provide an Digital Rights Management upgrade path to IPv6. Digital rights management is a comprehensive and flexible platform that Quality of Service enables rights holders to create business

The QoS function must include support The solution is applicable to not only for DSCP (Differentiated Services Code DOCSIS networks, but also other edge Point) as well as possibly other IP-based QoS networks including 3GPP, Wi-Fi, and mechanisms as specified by the operator. WiMAX wireless networks as well as going over the top on networks owned by third party Bandwidth Management service providers. The Bandwidth Management function will Functions for a core network are proposed monitor and police services and bandwidth that are not that different from cable networks usage. today. Content ingest remains essentially the same with the added ability to transcode Storage and Caching content to different codecs. Content distribution capabilities become more The Storage and Caching functions are sophisticated to replicate traffic around the used to keep select assets close to the edge. network where it is needed. Lastly, a group of Examples can be assets which are stored as an personalization functions are described which extension to the Distribution System allow cable services to be delivered to including movies or advertising content. customers in a way which matches their method used to attach to the network, whether Storage at the network edge is also used that be with a STB, mobile device, or some for remote storage DVR (Digital Video other customer premise equipment. Recording) in conjunction with the Consumption control plane. CONCLUSIONS

The paper presents a Video over DOCSIS solution using several available industry specifications including: • Modular CMTS

• DOCSIS 3.0

• PacketCable Multimedia

• Embedded DOCSIS

Modular CMTS in conjunction with DOCSIS 3.0 provide the most economical means to deliver linear, switched and On Demand digital programming over an HFC IP network. The proposed solution maintains backward compatibility with both existing DOCSIS 1.1 and DOCSIS 2.0 infrastructure and describes a method for delivering IPTV content to existing STBs. SWITCHED UNICAST VIA EDGE STATISTICAL MULTIPLEXING Ron Gutman, CTO & Co-Founder Imagine Communications

Abstract Video providers are unfolding their HDTV deployment plans: EchoStar carries Competitive HDTV offerings compel 30 National HDTV channels, plus locals, Cable Operators to make plans for 100+ HD planning on rapid growth. DirecTV plans to Broadcast channels and 10% HD-VOD ser- carry 60-100 HD by year’s end, with new vice (peak capacity), as well as devise inno- satellites providing capacity for 150 national vative in on-demand and Time Shifted televi- and over 1,000 local HD channels. Verizon sion services. Unfortunately, the current net- FiOS carry 20 HD channels, claiming “best works are unable to carry the load. The vari- video quality”. AT&T U-Verse plans to offer ous bandwidth enhancement options are 25 HDTV channels. capital intensive, disruptive to existing infra- structure and, for the most part, don't pro- It is apparent that the new competi- vide sufficient bandwidth. This study ana- tive landscape for acquiring and retaining lyzes SDV Unicast and Edge Statistical Mul- premium subscribers is centered on HDTV tiplexing (StatMux), concluding that it is the content. Cable operators must make plans for most cost-effective and least disruptive 100 HDTV broadcast channels and improved method for handling the HDTV bandwidth HD-VOD service. explosion. VOD has established itself as a useful “weapon” in this contest. There are 28 mil- HDTV-driven Bandwidth Explosion lion VOD-capable homes in U.S, with over 3

billion streams served in 2006, and a steady HDTV is beginning to permeate increase in adoption, usage and ARPU. mainstream America. Currently in over 19 TWC's “Start Over” service has been ob- million US homes, i.e. at 17% current pene- served to increase the on-demand usage by tration, HDTV is projected to climb to 81% 25%, and additional Time-Shifted TV ser- by 2010, according to Kagan Research. vices are emerging. Playlist and dynamic ad Unlike the historic migration from analog to insertion techniques are being tested to en- digital TV, whereby some percentage of the able a further increase in on-demand content subscriber base continued to access both and profitability. formats, it has been recently observed that a significant portion of HD tuners are tuned However, the spectral occupancy of almost exclusively to HD content, indicating the combined SD-VOD, HD-VOD and HD that once exposed to HDTV, some consum- Broadcast services is identified as a key lim- ers become disenchanted with SDTV. iting factor.

Concurrently, HD-DVD and Blu-ray Bandwidth Expansion Options are making their debut in the retail market, increasing the demand for HD content and How much bandwidth will be actually raising the video quality bar. HD-DVDs offer needed? Over the next couple of years, to a stunning viewing experience at average bit keep up with competition, cable operators rates (VBR) of up to 20Mbps using H.264 might be required to provide 100 HD Broad- compression. cast channels, 8-10% peak capacity HD- VOD service and up to 25% SD-VOD ser- particular, analog tier reclamation requires vice, to account for the increase in VOD li- many additional expensive STBs and high brary content, better movie release windows, installation costs. Moreover, even when the ad-supported content and Time-Shifted TV STBs are burdened on the cable operator’s offerings. Advanced high-quality encoders balance sheets, it has been observed that such and new post-encoding techniques will en- upgrades encounter significant resistance able multiplexing of three HD streams per from certain subscribers. Unlike the HDTV QAM channel (3:1) at good quality. A experience recounted earlier, some people do straightforward estimate indicates that the prefer analog (and no set-top) over digital operators will require 33 6MHz slots or about SDTV. Similar problems arise upon consid- 200MHz to provide 100 HD channels. On- ering upgrades to H.264, denser QAM, and demand usage will call for minimizing the 1GHz, all of which require STB replacement Service Group (SG) size at the fiber node, and expensive installation at the subscribers' down to about 500 digital tuners per SG. It is premises. There are other options such as estimated that at 20% HD penetration and spectrum overlay, FTTH and dense home 10% HD-VOD service, each SG must supply base decoder, which do not entail STB up- 10 HD-VOD concurrent streams. SD-VOD grades, however those require expensive usage of up to 25% implies 125 SD-VOD plant upgrades and/or expensive CPE home stream capacity. Together with HD-VOD, a gateways, and even higher subscribers prem- total of (10 x 4 + 125) / 10 = 17 QAM chan- ises installation expenses. nels per SG. This is 13 more VOD QAM channels relative to the currently deployed 4 Switched Digital Video channels per SG! Altogether, HDTV digital broadcast and VOD expansion will further A promising bandwidth expansion require about 40-50 additional 6MHz slots, option is Switched Digital Video. The pri- or 240-300 MHz. mary architectural alternatives for SDV are multicast and unicast. While multiple sub- The primary question is whether it is scribers can share a multicast stream, a uni- even physically feasible to provision the re- cast stream is unique to each subscriber. Uni- quired bandwidth. The second question in cast SDV enables more personalized media line concerns the economics: cost per sub- services, included virtually limitless content scriber per 6MHz slot. offerings and highly targeted ads, closely re- sembling Internet content delivery in its abil- In principle there are several band- ity to target specific content to specific con- width expansion options. We first consider a sumers. Most importantly, Standard Defini- costly plant upgrade from 750MHz to tion SDV Unicast can eventually include the 860MHz plant – unfortunately such an up- entire SDTV Broadcast tier, freeing up a lot grade barely provides an additional 100MHz of spectral space for HDTV Broadcast and bandwidth, hardly addressing the requisite HD-VOD. capacity enhancement. Node split is again expensive and only doubles the Switched tier but not the Broadcast tier, freeing at most SD Multicast two to six (2-6) 6MHz slots. A typical deployment of Standard Other bandwidth expansion options Definition SDV Multicast involves an SDV involve massive CPE upgrades. However, pool of 120-150 SDTV channels over 8 these options are too disruptive, too expen- QAM channels. In a Broadcast tier, assuming sive, and entail a certain amount of churn. In good-quality encoding, post-encoding and StatMux technologies, it would be few more QAM channels, but this is hardly possible to multiplex 15 SDTV streams per sufficient. Following is typical network ar- QAM channel (15:1) at today's typical SDTV chitecture with Standard Definition SDV quality. This means that 120-150 SDTV Multicast. channels in the Broadcast tier would occupy 8-10 QAM channels. Equivalently stated, The network architecture illustrated Standard Definition SDV Multicast savings, in Figure 1 provides 4 VOD QAM channels, in the way it is deployed today, would only 8 SDV Multicast channels and 32-42 Broad- amount to, at best, 2 QAM channels. By cast QAM channels depending on analog tier adopting a more aggressive content-to- size (72-80 channels typically) and the plant capacity ratio (sometimes referred as concen- bandwidth (750MHz or 860MHz). Assume tration ratio) it might be possible to save a typical Service Groups (SG) of 500-1000 homes-passed and 500 tuners, 400 of which service without adding a single QAM device are SD, and 100 HD at 20% HD penetration. while laying the groundwork for a robust fu- Under these assumptions, the architecture of ture bandwidth optimization roadmap; (ii) Figure 1 is able to provision 400-450 SDTV significantly improved video quality enabling Broadcast channels, around 20 HDTV more effective competition with less con- Broadcast Channels, only 8% SD-VOD ser- strained media such as HD-DVD/Blu-Ray vice and no HD-VOD service. and Verizon FiOS; (iii) a technology path to truly personalized ad insertion; and (iv) an SD Unicast & Edge StatMux improved overall user experience, in particu- lar instantaneous channel change support. Standard Definition SDV Unicast (SD Unicast), combined with advanced Edge In the proposed network diagram il- StatMux technology can address the band- lustrated in Figure 2 below, a dense Edge width issue in a cost-effective way and pro- StatMux either replaces the IP Switch for vide a non-disruptive migration. SD Unicast, VOD and SDV or is positioned immediately powered with Edge StatMux capability, pro- downstream from the IP Switch. The Edge vides: (i) limitless SDTV content, 100+ HD StatMux module receives SPTS/UDP/IP sig- Broadcast channels and 8-10% HD-VOD naling from the VOD Servers and over Multi cast IP (for the Switched and Broadcast sonalized/targeted ad insertion; instanta- tiers). The Edge StatMux outputs neous channel change; easy migration to MPTS/UDP/IP to the QAM devices, and op- Switched Unicast; and enabling content tionally locally multiplexes the Broadcast tier mobilization between Broadcast and as well as providing local ad insertion and Switched tiers. Notwithstanding these the ability to move content dynamically be- benefits, in order to actually provide a tween Switched and Broadcast tiers. The scalable solution, the bottom line should Edge StatMux is connected in a one-to-one be that the Edge StatMux price be below correspondence with the QAM channels at that of the conventional QAM evolving the QAM device, together forming the so- price, on a per stream basis. called "Edge QAM." An Edge Resource Manager (ERM) or a Global Session Re- 2. Video Quality. Under peak usage, source Manager (GSRM) communicates di- the Edge StatMux must provision at least rectly with the Edge StatMux in such a de- 15 SD streams or 3 HD streams per ployment, instead of communicating with the QAM channel, at the common video QAM device, for the purpose of signaling quality acceptable for these services to- channel changes. day. However, in VOD and Standard Definition SDV Unicast, most of the Edge StatMux Requirements QAM channels are under-utilized most of the time. At least 99% of time, the video The following list describes the es- quality threshold can be significantly ex- sential set of performance requirements and ceeded. During the brief times of peak features to be supported by an Edge StatMux, usage, the quality should nevertheless be under the proposed solution: preserved at a minimum threshold. The system should then comprise QoS 1. Price. The benefits of an Edge Stat- mechanisms ensuring uniform video Mux are multiple: substantial savings in quality under all conditions (detailed be- QAM device capital expenditures; free- low). ing up of 6MHz spectrum slots; signifi- cant improvement in video quality (to be further discussed below); enabling per- 3. Density. A single platform should support instantaneous channel change to support an entire hub of 50 or more Ser- further reduce the delay by another vice Groups, i.e., at least 600-1000 QAM 250msec on average. channels. The real estate utilization of hub's rack should not be increased as a 8. Encryption. It is usually preferable result of deploying the new devices. The in SDV deployments to use the more size of the platform should approximate cost-effective upstream bulk-encryption the size of the QAM devices being re- technology rather than session-based en- placed, and not to exceed 5-6 rack units. cryption. Some amount of VOD content is also being pre-encrypted at content 4. IP Switching. The Edge StatMux origination. Therefore, an Edge StatMux must support standard 1GbE fiber or must support the variety of encryption copper interfaces, and 10GbE in the fu- schemes: SDV bulk-encryption, VOD ture. It must perform non-blocking IP pre-encryption as well as session-based Multicast duplications, and supply a 2:1 encryption. IP unicast input to output ratio for ad in- sertion and VOD StatMux compression. 9. Management. Must support standard NGOD and ISA element management 5. Reliability. The Edge StatMux plat- and session control protocols, managing form must be Carrier-Grade with “5 by proxy the QAM device and the 1:1 Nines” availability, entailing No-Single- StatMux to QAM connection, as well as Point-of-Failure (No SPOF), redundant aggregate alarms and events. hot-swap fans, redundant load-sharing and hot-swap power-supply units, redun- 10. Splicing. Last but not least, an Edge dant hot-swap platform management StatMux should support the insertion of card, and hot-swap N+1 process- unique ads per stream with no capacity or ing/multiplexing cards. quality degradation.

6. Interoperability. The Edge StatMux Content must plug seamlessly into the existing in- frastructure. It has to be able to re-use the Standard Definition SDV Unicast, current installed base of broadcast, VOD combined with Edge StatMux capability, can and SDV QAM devices, as well as off- provide 100+ HD Broadcast channels, 8-10% load advanced functionality from the HD VOD service and unlimited SD content QAM devices. It should also support the for Broadcast, VOD and Time Shifted TV, installed base of VOD servers and possi- without adding a single QAM device (!). It bly off-load dynamic ad insertion and also enables controlled content migration of slow-motion trick mode functionalities. SD Broadcast to SD Unicast and service mi- gration, SG to SG and Hub to Hub. 7. Very Low Latency. The delay is not to exceed an additional 100msec in VOD Let's look at some usage assumptions trick-mode transitions or SDV channel and results: change. The Edge StatMux should also streams at the Broadcast tier or letting tuners tune to analog under peak usage will readily ensure the 40% peak usage assumption of SD Unicast.

ƒ Although it has been observed that HD tuners predominantly “want” to tune to HD content, since the HD content of- fering is still insufficient in 2007, it is as- sumed this year that users will still tune to substantial SD content over prime time and, and this is accounted for in the SD Assuming: Unicast capacity calculation. At the point where 100+ HD Broadcast channels and ƒ Typical Service Groups (SG) of 500- 10% HD VOD service are deployed, HD 1000 home passed and 500 digital tuners tuners are excluded from the SD Unicast out of which 400 are SD tuners and 100 calculation. are HD tuners (20% HD penetration in 2007). ƒ SD VOD, Time Shifted TV, nPVR services are in high demand. In SD Uni- ƒ Aggressive digital penetration of 10% cast, these services are unlimited. Over more digital STB per year. time, similar services will be required for HD and therefore it is assumed that HD- ƒ HD penetration growth projected to VOD peak usage is to grow at a rate even climb to 81% in 2010 according to Kagan faster than the SD-VOD growth curve. Research. The results of this analysis are quite Regarding the next two numbers to be as- promising. With just 12 QAM channels per sumed, it is important note that network de- SG it would be possible to deliver unlimited sign should be based on peak usage. The SD content and 8-10% HD- VOD service, peak video consumption occurs Thursday while freeing up the spectral resource for evening Prime Time, 8-11PM. The number about 100 HD Broadcast channels. In the of tuners that are on in the Sun-Wed window 2009-2010 timeframe, if HD penetration in- is typically around 20%, however, on Thurs- creases as projected, a SG split of 2:1 will be day evening it might peak to 60-70%. There- further required to maintain 12 QAM chan- fore: nels per SG, in order to keep up with the in- creased HD-VOD services. ƒ Maximum number tuners on is 75% per SG, SD or HD. In the future, assuming SG/node splits will eventually reduce the number of HD ƒ Maximum tuners watching digital SD tuners per SG to 250-300 HD tuners, this will at peak usage is 40%. On Thursday night, enable, in conjunction with the Edge Stat- it is estimates that 60-80% of subscribers Mux architecture further expansion to unlim- are tuned to 20-30 channels in the ana- ited HDTV SDV Unicast, completely remov- log/simulcast tier (e.g., ABC, CBS, NBC, ing the bandwidth constraint. At this point, Fox, CW, USA, FX, Fox News, ESPN, the age-old adage “never enough bandwidth: CNN, MTV, Disney, HBO, Discovery, may finally be inapplicable to cable’s robust PBS, etc.). Dynamically leaving 60 SD architecture. Video Quality Management quality sports channel peaks at about 5Mbps but may averages approximately 2.5Mbps. Video compression standards such as Applying multi-rate CBR coding, it is only MPEG-2 and H.264 take advantage of tem- possible to carry 8 such “difficult” services poral redundancies. MPEG-2 encoding of an over a 38.8Mbps 256QAM channel. In con- SDTV signal, which is temporarily rich in trast, with VBR/StatMux at the same quality, motion and scene changes takes 4-4.5Mbps, it would be possible to carry 15 channels, i.e. providing the same quality as a "talking- almost twice as much. heads" scene at 2Mbps. This is why a consis- tent quality coding technique must use the In fact the process of allocating bit Variable Bit Rate (VBR) format, which pro- rate by the ERM/GSRM in a VBR/StatMux vides the optimal video compression scheme. is akin to allocating bit rates in a multi-rate This is also why DVDs use VBR to optimize CBR system: every stream is assigned a dif- video quality and storage efficiency. More- ferent effective bit rate which resembles its over, virtually all digital broadcast multi- VBR average, e.g. a sports channel may have channels content in North America uses VBR a higher average bit rate than a news channel. and Statistical Multiplexing. The Staging Processor continuously monitors the VBR average and publishes it to the ERM/GSRM. The average bit rate is meas- ured statistically per channel and may attain multiple values according to the time of day/week. A video quality monitoring system analyzes the VOD assets in advance in a similar manner, providing a deterministic VBR average to the ERM/GSRM over ADI

interface bit rate field. Significantly, CBR encoding or the

"clamping" process of conversion from VBR Until now it has not been economical to CBR, limit the video quality by chopping to extend VBR usage to the video distribu- off the peaks. Moreover CBR multiplexing tion network edge for advanced service archi- reduces the QAM and bandwidth efficiency tectures such as VOD and SDV. However, by another 33% down to 10 streams per the Edge-StatMux architecture changes the QAM channel. equation. The new VBR StatMux solution

optimizes video quality over each QAM channel. Advanced Edge StatMux technol- ogy can multiplex 15 SDTV streams at broadcast quality, without sacrificing the economics. In VOD and switched video en- vironments, since the QAM channels are not fully utilized 99% of the time, it is possible to further raise the quality bar without sacri- Current advanced SDV systems plan ficing the capacity gains attainable with the to overcome the CBR quality degradation Edge StatMux architecture. problem by using a method called "Multi-rate CBR", assigning higher bit rates for selected high-complexity services such as sports channels, therefore reducing the QAM effi- ciency even more. With VBR coding, a good Conclusions

In this paper we addressed the various options to mitigate the bandwidth demand induced by the accelerated HDTV and VOD penetration trends. It is concluded that a promising avenue to meet this challenging spectral demand is a Switched Digital Video variant involving Standard Definition SDV Unicast and Edge Statistical Multiplexing (Edge StatMux). The benefits of the architec- ture in terms of bandwidth efficiency, de- ployment migration, video-quality, and over- all user experience (including personalized ad insertion) are coupled with an upward leap in video quality enabled by porting the VBR video format to the network edge, with in- creased video quality and bandwidth effi- ciency. Our deployment projection model accounting for content and usage patterns indicates that the new architecture provides an important new tool in the arsenal of the cable television operators, to economically and competitively manage the HDTV and VOD penetration trends, while meeting en- hanced quality of service requirements.

[email protected]

THE ‘BRIGHT SIDE’ OF DRM: NEW BUSINESS OPPORTUNITIES Robert M. Zitter and Craig D. Cuttner Home Box Office, Inc.

Disclaimer. The potential business scenarios described herein are outlined for technical feasibility and do not represent any proposed or actual business offerings.

• Rules for permitted copies that can be Abstract made on certain devices under controlled circumstances Digital Rights Management (DRM) has been, unfortunately, characterized as an • Rules for content movement and “evil” method of restricting user’s rights and requirements for protected outputs or abilities to consume content. Quite the security of devices including contrary is true – the use of DRM, providing downstream (subsequent) other DRM usage rules and advanced security, enables systems. new opportunities and new business offerings. This paper describes some of the Different DRM systems may have potential business offerings that might be additional features or slightly different provided if an operator were to use DRM to implementations but for the purposes of this distribute content outside the normal discussion, these core features will be used to scenario of linear content and traditional build business models. VOD. As an example of how varied DRM INTRODUCTION system capabilities can be, here is a summary of a few key features of major DRM systems Digital Rights Management (DRM) from public sources: represents a comprehensive system of cryptographic controls tied to business rules that govern the use of content. Basically, DRM is a combination of Conditional Access (CA) that is used as an on/off switch for payment, and Copy Protection (CP) the coarse control over the duplication of content.

DRM adds the component of usage and business rules to the consumption of content – examples of content rules:

• Defined consumption (playback rules) such as number of permitted viewings, time limit before viewing window ends;

DRM Feature Microsoft Microsoft DRM Apple Fair Play Cable CA & CCI Restrict Access to “content for payment” Yes Yes Yes Limit copying to Free / Once / No Copies Yes Yes Yes Can require protected outputs (e.g. Macrovision) Yes No Yes Limit unprotected outputs Yes Yes Yes Limit viewing time and copy retention Yes No No Permit / Restrict movement specifically to portable Yes Yes No devices

NEW OPPORTUNITIES Newer disc formats (e.g. HD-DVD) have an advanced system of copy Existing “sales” of content generally management (Advanced Access Content reflect either a single narrow use (e.g. VOD / System (AACS)) that is more of a full DRM PPV) or almost a perpetual right to view system to address this shortcoming. content (such as physical ownership of a However, business plans that wish to target DVD). Intermediate variations are also the millions of existing (CSS-only) DVD possible such as limited-use subscription, or players are obliged to “regress” to that transaction-based rental. technology.

Most opportunities that use DRM Since DVD content is “authored” require that all “devices” in the chain of (both MPEG formatting as well as menus, content and/or transaction have an chapters and other non-cable formatted data), implementation of the same DRM system it is impossible to “capture” a DVD from a and/or a compatible downstream DRM live streamed program and burn to a disc. system via an approved output – including Content to be burned to disc has to be the “client” device used to consume the distributed to the burning device (e.g. content. specialized STB) using DRM to encapsulate the CSS-encoded DVD image file.

Scenario One – “Burn to Own” DVDs Scenario Two – “Portable Devices”

Conventional (standard definition) Currently, conventional “linear DVD’s use a copy-prevention system called broadcasts” are “fat” – i.e. they are Content Scramble System (CSS). Other than distributed for pretty pictures on big screens the prevention of copies (and, in some cases, – this is a poor choice for portable devices enabling output controls such as with small streams and small disk drives. Macrovision), there are no other proscribed Thus, a content “move” from DVR to DRM attributes such as limited portable device is inefficient and expectation “permissions” of copying. of transcoding (bit-rate reduction) is a burden for existing DVR STBs.. that do not support Macrovision, limiting, An improved method to provide perhaps this offering to standard-definition content to portable devices is to encode content as copy protection is not specifically for that target device and implemented on high-definition analog efficiently move content securely to the outputs. In these cases, either digital outputs portable device. with HDCP or image constraint would be required. Unprotected outputs would Secure DRM can be used to authorize severely increase the risk that analog-sourced the delivery of content to portable devices. copies could be made and undermine the very significant home video-window revenue stream. Scenario Three – “Electronic Sell Through” Another “Early Window” content idea Without advanced DRM, a file represents a potentially new revenue stream transfer of content to a PC would be that may afford electronic distributors with a considered “lost” to the Internet, thus any very visible (and promotable) offering to “retail” transactions that enable consumers to consumers. With extremely robust DRM and have and use content require a sophisticated content controls, content could be offered DRM. much closer to the theatrical “first-release” than was previously possible. Movie titles EST is the process where a “locked” (for example), could be offered on a very file is distributed to a user with a proscribed restricted and limited basis to HDTV Home rights-enabled DRM license to permit Theater households very near their release to viewing and/or movement to portable theaters. Of course, copying or other content devices and inside the home (e.g. on a home leaks could undermine several subsequent network). revenue streams and would not be tolerated. Thus, an early window release could include some or all of the following restrictions Scenario Four – “Early Window High Value (which, due to the nature of a new business Content” offering like this, may require review of the FCC’s rules that govern output controls, Two variants of potentially- etc.): interesting new business opportunities are possible with a highly-secure and usage- • Mandatory highly-secure distribution restricted environment. environment

First, “Day and Date” refers to the • Copy Never on protected digital deployment of on-demand assets that are outputs, perhaps restricted to HDCP available to consumers coincident with their and not Firewire home video consumer release. This represents an interesting opportunity since • Analog unprotected outputs disabled the consumer interest and revenue model of (Selectable Output Controls (SOC)) the existing home video marketplace are well characterized. Such distribution would • Limited viewing window (e.g. no probably require both copy protection (copy pause, no 24-hour bookmark / never, Macrovision protection for analog retention) outputs) and potentially, restrictions on other outputs – such as component analog outputs • HD Only distribution (since this offering may be potentially targeted to home theater owners)

CONCLUSION

The proper end-to-end implementation of advanced DRM involves cooperation of distributors, manufacturers (CE and others) as well as assurance that the overall robustness of the distribution architecture is without question.

While the form of new product offerings and revenue opportunities will be determined by those who create and provide them, their existence will all rely upon one common denominator: The use of a robust Digital Rights Management System capable of attaching and enforcing the specific rights granted to consumers who use these new products. Our ability to manage the digital rights associated with the video programming will open the door to new choices for consumers and new revenue for programming distributors. THE COST OF FAIR BANDWIDTH: BUSINESS CASE ON THE RISE OF ONLINE VIDEO, P2P, AND OVER-THE-TOP; PROACTIVE STEPS OPERATORS CAN TAKE TODAY TO MEET BANDWIDTH DEMANDS IN THE NEAR FUTURE Michael Eagles, UPC Broadband Robert F. Cruickshank III, C-COR

Abstract product speeds. We discuss “consumption” in the context of usage trends and subscriber The growth of popular video sharing “bit-rates” or average speed in Kilobits per sites such as YouTube, P2P file sharing second (Kbps). applications such as Bit Torrent, and over- the-top broadband services has placed Why Look at High Speed Data? unprecedented strain upon the HFC network. Demand for the broadband pipe is growing High Speed data is one of the few daily and in order to ensure that all services services where forces, largely external to the and applications are delivered reliably, cable operator, shape the demand for capacity. operators must plan for the traffic increase In the recent rise of online video, P2P and and take proactive steps to mitigate the rising over the top multimedia services, we notice a demands. trend from the use of High Speed Data for “access” toward its use for “entertainment”. This paper will take a multi-pronged view of In this paper we seek to understand the impact the demands placed on today's network, of this shift on bandwidth needs. including a recent history traffic analysis of sample systems and projections of where Methodology networks are headed tomorrow. Additionally, the paper will investigate both hardware and We considered the level of congestion software options for proactively working to on today’s network by looking at a snapshot meet bandwidth demands today and over the of global consumption patterns in addition to next three to five years. a specific per system example.

INTRODUCTION In assessing the recent and projected growth in demand and bandwidth needs we Increasing Bandwidth Needs considered both historical trends in per subscriber average capacity, per subscriber With operators launching many profiles and device and application trends. bandwidth intensive services including High Definition Television (HDTV), Video on Finally we considered a tool-kit of Demand (VOD) and ever-increasing High alternatives for the cable operator, grouped Speed Data rates, there is an ongoing need for into “manufactured bandwidth” or capacity greater capacity. Here we take a close look at expansion and “technology bandwidth” or High Speed Data and the rise of online capacity management. multimedia devices and applications to better understand current and future demand trends.

In this paper we use the term “bandwidth” to mean “RF bandwidth” and “information rates” to mean advertised

AN ANALYSIS OF TODAY'S NETWORK 1995 in a major metropolitan city in the DEMANDS United States.

In this first section of the paper we explore a snapshot analysis of the demands placed upon the HFC network by identifying global trends and considering data services in a major metropolitan city in the United States.

Average Usage per Subscriber

We started with a review of global usage trends using a base of 2.75 million DOCSIS® Cable Modem (CM) devices. As shown in Table 1, the average high speed subscriber sends/receives about 300 Megabytes (MB) of IP traffic per day. As one might expect, Figure 1. Average Bandwidth Cumulative average daily usage varies among different Distribution in June 2005 regions of the world, but not drastically. Referencing Figure 1, one can see that Average Daily Directional Usage [MB] just 3% of the CMs (on horizontal axis) sent Region Avg. Up Avg. Avg. 58% of upstream traffic (on vertical axis in Down Up+Down USA 108 210 318 blue) and received 47% of the downstream W Europe 123 192 315 traffic (on vertical axis in red).! E Europe 95 183 278 S America 138 176 314 Fast-forwarding almost two years to Asia 98 190 288 today, we see similar results in the upstream—just 3% of the CMs sent 56% of Table 1. Typical Average Daily CM Usage upstream traffic. But look, over the same time i by Region during November 2006 period the downstream number has dropped from 47% to 37% as shown in Figure 2. In order to identify sources of congestion on today’s network we take a closer look at how this average usage is distributed across the user base and what attributes contribute to this usage.

Usage by “Power Users”

A common belief is that some subscribers send/receive more or less traffic than other subscribers. A logical follow-on question is “What percentage of subscribers send/receive more than others?” Figure 1 shows the cumulative percentage of traffic Figure 2. Average Bandwidth Cumulative sent/received by ~500,000 CMs during June Distribution March 2007

The takeaway messages from Figures 1 Notice in Figure 3 that the percentage of & 2 are: a) half to three-quarters of all IP active devices never falls below 31%; put traffic is sent/received by just 5% or less another way, about one-third of all CMs subscribers, and the remaining 95% of actively send/receive traffic at all times. subscribers send/receive the balance of IP traffic on MSO DOCSIS® networks; and, b) Zooming out to a whole year in Figure the distribution of volume has changed due to 4, one can see the percentage of active devices an increase in downstream consumption by grew more than 10% in 12 months. The the “average” user. takeaway message is that over time a greater percentage of the overall CM population We believe a possible reason for the actively send/receive traffic (are active) distri bution change is that bandwidth around-the-clock. We discuss possible intensive, multi-media applications are reasons for this in a later section. becoming widely used by mainstream subscribers rather than only early adopters. Fast-forwarding almost two years to Active vs. Total Cable Modems today, we see results quite similar to the trends shown in figures 3 and 4. Another question is “What percentage of CMs are active throughout the day and night?” Figure 3 shows the number of active and number of total DOCSIS devices (CMs) on a typical Metro upstream hour by hour over 1 week. The total number of devices (across the top) is relatively consistent and ranges from 216 to 220 (the number varying as CMs become unreachable, for example, when subscribers shut down/power up their CMs). The active number of devices (lower periodic trace) varies throughout day from 63 to 148. Dividing the number of active devices Figure 4. Active & Total Devices on a by the number of total devices at any hour Typical Metro Upstream in 1 year yields the % of active devices (upper periodic trace) which is read on the right-hand vertical axis. Rise in Measured Congestion

A greater percentage of CMs send/receive more IP traffic more of the time, resulting in a rise in HFC congestion. The top half of Figure 5 shows the hour-by-hour rise and fall of congestion across all of Metro’s 500,000 CMs over one week. Notice that peak congestion occurs daily at 8-10 PM. The

highest congestion level was late Sunday Figure 3. Active & Total Devices on a night when about 40,000 of the 500,000 Typical Metro Upstream in 1 week Metro CMs experienced slowdowns due to congestion.

caused by new applications and devices that create/consume IP traffic.

RECENT AND PROJECTED GROWTH OF THESE DEMANDS

In this section of the paper we explore a historical view of the increase in information rates and the impact of new multimedia online services such as YouTube. By exploring recent growth, we will extrapolate the magnitude of traffic that can be expected within the next three to five years for Figure 5. Weekly & Yearly effects of residential broadband data. Congestion across all of Metro The Digital Household The bottom half of Figure 5 shows the day-by-day rise and fall of congestion across We believe there are two primary factors all of Metro’s 500,000 CMs over an entire at work contributing to the increase in year. The vertical axis now represents the measured congestion on HFC networks. number of modem congestion hours (the sum of the number of CMs congested per hour for (1) Growth in Information Rates: Using all 24 hours in the day). Over the course of the historical trends as a base-line we note year congestion levels rose nearly 600% from continued increase in information rates that an average of ~50,000 modem hours per day lead to an increase in measured congestion. to ~300,000 modem hours per day. This is fueled by the competitive nature of the broadband access market. What is the Relationship between Congestion and Bandwidth? (2) Growth in Multimedia Applications and Connected Devices: New multimedia Even when it is assumed that the applications, including embedded multimedia, traffic characteristics do not change, an P2P television, and TV place-shifting, result increase in the available bandwidth to the in more bits being transported. In addition, subscriber - the “information rate” - results in wireless home gateways are entering the peak rate requirement increases. As a result, mainstream, and we see growth in the number the of IP-connected, traffic-generating devices network can serve fewer subscribers at the behind each CM. same level of congestion or the same subscribers with increased congestion. We see these two factors working together to create significant additional demand for In considering the recent and projected network capacity. growth in demand, we consider both historical and projected information rate increases that contribute to congestion as the speed of the sources (CM’s) increases. In addition, we consider the impact on traffic characteristics

Predicted Information Rates 2006 2008 2010 (Kbps) (Kbps) (Kbps) To predict the recent and projected growth High 20,000 60,000 160,000 of demands on the HFC network we also Case considered an analysis by Bob Scott ii that Probable 10,000 20,000 60,000 extrapolates historical growth in data product Case advertised speeds since 1982. This analysis Low 7,000 10,000 20,000 notes a close fit to Moore’s Law and provides Case the following high speed data product Weighted 8,000 14,000 30,000 information rate scenarios shown in Figure 6. Average

Using this model we start in 1982 with Table 2: Predicted Information Rates in Kbps dial-up telephone Modem speeds of 300 bits per second (bps) and climb to nearly ~20 megabits per second (Mbps) today. We expect this increase in CM information rates to increase congestion and

1 5 1000000 7 ,2 8 9 5 6 4 8 3 required bandwidth even if the bits ,9 3 ,7 8 1 1 2 4 8 1 , 9 5 100000 , 7 6 3 6 9 1 ,8 3 transported and the number of CMs stays the 4 0 3 ,9 ,6 1 8 5 2 10000 6 2 0 1 0 ,2 0 9 2 9 same between 2006 and 2010; and we expect 2 5 2 0 0 0 3 1000 0 7

1 1 5 0 4 0 the impact on required bandwidth and 0 7 5 7 1 2 2 3 5 100 8 6 1 2 1 8 4 congestion to be higher in the case that the 5 6 3 2 3 8 10 4 1 4 9 . . 4 bits transported increases through new 6 1

Information Rate (Kbps)Information 2 1 .4 0 1 .2 applications and additional devices per CM.

0 0 .3 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 9 9 9 9 9 9 9 9 9 0 0 0 0 0 0 0 0 0 8 8 8 9 9 9 9 9 9 0 0 0 0 0 0 1 1 1 2 5 7 0 2 4 5 7 9 0 2 3 4 6 8 0 2 4 Demand Model and Usage Profile Moore's Law Prediction Actual PIR

Figure 6. Moore’s Law Growth in Historical An analysis of traffic profiles described by and Predicted Information Rates using John T. Chapman’s Multimedia Traffic Engineering Model published in 2002 A sensitivity analysis from this work indicates 1 to 2 traffic generating devices per CM iii. Recent research suggests that this has indicates a range of information rate or iv “speed” possibilities as shown in Table 2. We increased to over 3 devices . assume a subscriber take-rate of the different speeds in Table 2 - High Case of 5%, We assume an increase from 1.5 devices Probable Case of 10%, and Low Case of 85% per CM in 2006 to 3 devices per CM by 2010 - reflecting a mix of subscribers using explained by increased market usage for new different broadband product speeds to create a broadband connected applications. weighted information rate each year. In addition to considering historical per subscriber Kbps data we considered traffic contributions from existing applications on 1.5 connected devices per CM plus contributions from an additional 1.5 connected devices supporting several new applications including Embedded Multimedia (i.e. YouTube); P2P Television (i.e. Joost); VOIP Applications (i.e. Skype); TV

Placeshifting (i.e. Slingbox), and CE Devices (i.e. Connected Handycams).

Using this traffic engineering model in conjunction with recent new applications and devices we are able to estimate the future application traffic in average Kbps per data subscriber from approximately 80 Kbps in 2006 to 350Kbps in 2010 as shown in Figure 7.

Kbps Per Subscriber Trends and Forecast Figure 8. Internet Protocol Trends vi 1000 350Kbps 1993 1999 2006 2010 ? 80Kbps 100 FTP Web P2P P2P-TV New multi-media applications & 23Kbps Email FTP Web Rich-Web increasing number of devices in the vii household increase the Web Email News P2P-Other 9Kbps amount of congestion 10 … leading to continued increase in Kbps per Other Other Other Other subscriber. Average Kbps Per Subscriber Table 3. Internet Protocol Trends viii 1 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 To understand two possible dominant Figure 7. Kbps Per Subscriber Trends v applications we explore the growth of Embedded Multimedia Traffic and PC-Based Television and P2P Multimedia which in our We look in more detail below at changes in model contribute a significant amount of the the application mix over time to support this. new traffic.

Changes in Application Mix over Time Embedded Multimedia Traffic:

We asked the question “What application Recent developments in web applications changes are likely considering historical include the addition of video-based embedded trends?” Based on information illustrated in multimedia. Examples include Brightcove, Figure 8 we believe that P2P in its current Coull, Google Video, MediaZone, MetaCafe, ‘download’ form may have reached its peak PermissionTV, Revver, VideoJug, YouTube, and that a ‘TV centric’ and new multimedia Yahoo! Video, and Ziddio. We explore web applications may provide the next wave YouTube in more detail for this category. of traffic. Google's YouTube founded in Feb 2005 ix Considering the rapid rise of P2P traffic, provides flash-based video clips embedded in we note that the primary traffic generating a web browser. Measurement of Google’s application may not even exist at the time the YouTube traffic in Europe indicates that network is being planned! The overall mix YouTube represented 2.5% of all downstream and contribution of various traffic-generating traffic year end 2006 x, and growth of applications rises and falls over time as shown YouTube traffic was 400% for the 6 months in Figure 8 and Table 3. ending 2006. Normalized for subscriber growth this translates into 1 Kbps per subscriber today Swoosh, and Rawflow xii. growing to over 100 Kbps per subscriber in just 12 months. Is growth as shown in Figure We note that many P2P applications 9 realistic? generate traffic from a series of machine queued requests. Machine-to-machine traffic, YouTube: Kbps Per Subscriber Growth Forecast 12 months as the name suggest, indicates that the 1000.00 application usage continues even without 100.00 human activity (i.e. mouse-click) present as shown in Figure 10. Perhaps this helps 10.00 explain the rise in active modems shown in

1.00

(Logarithmic) Figures 3 & 4.

0.10 PeakHour Kbps Per Subscriber Web Multimedia P2P Multimedia

0.01 Jul-07 Jul-06 Apr-07 Oct-07 Oct-06 Jan-08 Jun-07 Jan-07 Jun-06 Mar-07 Feb-07 Sep-07 Nov-07 Dec-07 Sep-06 Nov-06 Dec-06 Aug-07 Aug-06 May-07 Date

Figure 9. Recent You Tube Traffic Growth

We believe that, more than likely, embedded multimedia traffic growth per subscriber will slow in the future based on reaching maximum market usage for a given broadband subscriber population. Figure 10. Web/P2P Multimedia Usage

Using the Chapman Multimedia Traffic Using the Chapman Multimedia Traffic Engineering single user profile, we assume Engineering single user profile we considered that by 2010 an increase in embedded that users watching television have a peak multimedia web content including adoption of time (on-peak) market usage of 60% and an Google’s Click-to-playxi advertising model, off-peak value of 33% of broadband results in an embedded multimedia user base subscribers by 2010 based on the Nielsen reaching market usage of 20% of broadband “Households Using Television Number” xiii. subscribers on average; and that the average We assume the average for P2P video is 40% bit rate is 250Kbps. This equates to 50Kbps and a blended average bit rate of ~550Kbps in of downstream traffic at maximum market the downstream and ~125 Kbps in the usage using our assumptions. upstream. This equates to 220Kbps per subscriber in the downstream and 50Kbps in PC-Based Television and P2P Multimedia the upstream.

Another category of growth in multimedia Today P2P video applications are PC- applications is PC-based Television and P2P based. However, in the future such an Multimedia. Recent developments include the application may find a low cost hardware introduction of several PC based multimedia “host” vehicle such as a set-top, thus online platforms including, Arvato, AOL’s significantly increasing market usage levels. In2TV, Babelgum, BitTorrent, Grid Networks, Kontiki, Itiva, Joost, JumpTV, MediaZone, Network Foundation Technologies (NFT), Peer Impact, Red We believe a future statistical projection of Joostxiv traffic would assist in identifying MHz) downstream port capacity in 2006 and growth patterns and a likely market usage 2010 is reserved for: a) other services (i.e. peak. eMTA based VoIP and STB return path), and b) additional growth. This results in 40Mbps Comparing Each Approach available for use per 8MHz channel. We also assume statistical multiplexing gains of 1.1 We considered both Moore’s Law from bonding channels. historical information rates, keeping congestion constant, and Chapman’s Multi- In 2006 we support 500 subscribers per Media Engineering model to create a single 8MHz downstream at 80Kbps per sub CM profile. compared to 2010 where we are able to support 100 subscribers per 8MHz Historical Single CM downstream due to the additional of new Information Profile using applications and new devices. To support the Rates and Chapman’s equivalent density per service group we Moore’s Law Model require 32MHz of downstream capacity and % Increase in 25.6MHz of upstream capacity as shown in 300% 341% Bit-rate 2006 Table 6. (320 Kbps) (353 Kbps) to 2010 Table 4: Comparing Each Approach 2006 2010 Number of DS 1 x 8MHz 4 x 8MHz For the purpose of determining capacity Number of US 2 x 3.2MHz 4 x 6.4MHz requirements we noted that Chapman’s Table 6. Additional Capacity Required model generated a greater increase in the per user bit-rate and was therefore used as the That’s 4 channels x 50 Mbps/ channel = basis for our 2010 assumptions. 200 Mbps or 32MHz of total capacity in the downstream as outlined in Figure 11 below. Predicted Capacity Requirements Subscribers Per 8MHz Downstream and Number of Downstreams

As a result of the application analysis 600 5 above we are able to estimate the amount of 500 additional future downstream and upstream 4 400 capacity that will be required. 3

300

Assuming the information rates offered by Number of 8MHz Dowsntreams 2 200 Subscribers Per Downstream Subscribers Per 8 MHz the Moore’s Law projection and the Kbps per Number of Downstreams 1 subscriber generated by new applications 100 using the Multimedia Engineering model, this 0 0 suggests the following over-booking factors 2006 2010 as shown in Table 5. Figure 11. Additional Downstream Capacity 2006 2010 Avg. Speed 8,000 30,000 Additionally we would require 60 Mbps Kbps 80 350 or 25MHz of upstream capacity as outlined in Over-booking 100 85 Figure 14 below. Subs/8MHz 517 (~500) 116 (~100) Table 5: Speed and Kbps per Subscriber We assume that 20% of the 50 Mbps (8

Subscribers Per 6.4MHz Upstream and Number of Upstreams Reduce Serving Group 1. Un-combine RF Nodes 2. RF Node split 600 8 Balance Users in Channel Line-up

3. Load balance across 5. Expand RF 500 4. Add RF Channel RF Channels Bandwidth to 1+ GHz 6 400 Increase Data Rate

6. Decrease FEC 7. Increase Modulation 8. Increase Channel 300 4 (US only) 16QAM – 2048QAM Width – Bonding Up-sell Subscribers 200

Subscribers Per 6.4 MHz 6.4 Per Subscribers Number of 6.4 MHz Upstreams 9. Bill subs for bandwidth consumed 10. Tier Data Rates

2 Upstreams MHz 6.4 of Number Subscribers Per 6.4 MHz Upstream 100 Reduce Subscriber Traffic 12. Selectively Rate Limit, but 11. Rate Limit Applications 0 0 on demand only as needed 2006 2010 Figure 13. Capacity Management Figure 12. Additional Upstream Capacity Tactics & Options

In summary, the growth in usage per Reduce Serving Group subscriber and changes in applications together reduces the net number of We observe there are two ways to subscribers per 8MHz downstream from 500 reduce the size of a serving group: Option 1 is to something more like 100 as shown in to Un-combine RF nodes so that fewer nodes Figure 12. (and subscribers) are connected to an upstream or downstream port. Option 2 is to MANUFACTURED BANDWIDTH - split a RF node into two or more serving EXPAND CAPACITY TO MEET DEMAND groups with fewer subscribers. In the limit, Option 2 would be considered a “Fiber Deep” Given the usage expected today and approach to reducing serving group size. within the next few years, how can operators prepare from an HFC perspective? This Balance Users across Channel Line-up section outlines the cost of various Capacity Management tactics including: As penetration grows it is safe to assume the existence of multiple channels. Option 3 (a) Reducing Serving Groups involves spreading users across multiple existing channels to evenly distribute traffic (b) Balancing Users across channel line-up loads. Option 4 involves adding additional channels (assuming spectrum is available), (c) Increasing Data rates and Option 5 requires expanding the HFC delivery infrastructure to accommodate the (d) Up-Selling Subscribers transmission of higher frequencies (e.g., 1GHz). (e) Managing Subscriber Traffic Increase Data Rates Each of the above tactics may have one or more options as shown in Figure 13. In DOCSIS systems, upstream traffic ‘packets’ may be ‘protected’ from packet errors by different levels of Forward Error Correction (FEC). In fact, it is possible to not use any FEC at all, though field experience dictates that some level of FEC should always be used. To successfully transport un-errored packets, clean portions of the HFC require P2P) around-the-clock, whereas an example less FEC and noisy portions require more of Option 11 is an on-demand as-needed rate FEC. While ensuring critical packets (such as limiting algorithm that enforces application or voice) are delivered unerrored, FEC comes at individual subscriber rate limits ONLY as a price; greater FEC (overhead) results in less needed (i.e., at the onset of congestion or User Data (payload). when congestion persists).

Clean HFC plants can also support The Cost of Capacity Expansion higher order modulation which means more bits (a larger alphabet) can be sent at a given We list the relative capital cost of some symbol rate. The take-home message in of the different capacity expansion tactics & Options 6 & 7 is that cleaner HFC carries options in Table 7. The important factor is more User Data than noisy HFC. that costs can vary dramatically between expansion options, and that a combination of Option 8 is the much talked about different options can be considered (i.e. solution known as channel bonding wherein Switched Digital Video & Bonded Channels). multiple channels are ‘bonded’ into a wider (super) channel with higher data rates and $ Per $ Per HP $ Per Data higher statistical multiplexing efficiencies Node Per at 2,000 Sub MHz of HP Per (25% pen.) Capacity Node Up-Sell Subscribers Upgrading to $714 $50 $200 1GHz xv Options 9 & 10 imply that ‘heavy’ users Fiber $107 $46 $184 who send/receive large amounts of traffic Deep xvi could/should pay more for service than ‘light’ Node $30 $13 $52 Split xvii users who don’t send/receive much traffic. Switched $312 $5 $20 The thought is that pricing can/will Digital xviii discourage heavy use. Option 9 could be a Bond $632 $18 $70 unit measure that can be thought of as ‘Bill by Channels xix the Byte Transferred’ which is akin to how many utilities/consumables are billed today Table 7. The Cost of Manufactured (e.g., milk, water, gasoline, etc., each cost so Bandwidth much per litre.) Option 10 is similar to Option 9, but instead of billing per byte, bills on TECHNOLOGY BANDWIDTH - MANAGE chunks of bytes (e.g., 0 ≤ 10 GB per month is SCARCE RESOURCES WITH NEW $40, 10 ≤ 20 GB per month is $50, etc.). TECHNOLOGY

Reduce Subscriber Traffic Given the projections, expansion approaches in isolation will not be enough to Options 11 & 12 refer to methods used maintain near-term demand. In addition, a to limit certain or all users’ traffic and/or software management approach to directing traffic types. Before getting alarmed at either and managing traffic on the network will Option, take a moment to consider a few stretch the increases gained with HFC practical realities. All highways slow down equipment upgrades, node splits, etc. This when they get full; Internet pipes slow down section outlines sample traffic management when they get full, too. Option 11 involves applications that ensure bandwidth is fairly rate limiting specific applications (such as provisioned so that high priority traffic such as VoIP calls are given top priority, while minimizing the impact of P2P and over-the- It is critical that the delayed application top applications. can be accurately identified and that a mechanism exists to detect a move by users to We see four possible instruments to alternative applications for the same use. (i.e. allocate scarce bandwidth: a move from a P2P protocol to Network News (NNTP) protocol when wanting to download (1) Rationing multimedia content).

(2) Delay Pricing

(3) Pricing Pricing may be used to impact consumption and may be based on volume, (4) Access nature of the service and across the day (as with the electricity industry). (5) Supply Application specific charges can also Rationing impact traffic consumption. For example, it may be possible to offer certain P2P services Operators are able to limit volume and within the monthly broadband service fee and access to certain applications. This can be to charge for access to others. hard rationing or soft (e.g. first warn heavy users and if behavior doesn’t change, then Alternatively it may be possible to deliver a penalty). charge a carriage fee for premium P2P content providers in return for a fair allocation of For example, dynamically re- bandwidth. provisioning a subscriber at a lower speed when consumption reaches a pre-set limit is a We believe that pricing is one of the possible way of making additional bandwidth most effective mechanisms for managing available to remaining subscribers. heavy bandwidth consumption.

Rationing has the disadvantage of Access penalizing all applications even those provided by the operators such as email, and With the growth of wireless home also penalizes the subscriber for heavy off- access points, theft-of-service can lead to peak consumption that may not impact increased traffic levels associated with a operator capacity at peak time. subscriber, even without the knowledge of the subscriber. Offering a managed home Delay network enables the operator to reduce the risk of theft and improves the security offered It is possible to delay the delivery of to the end user. non-revenue generating consumption though the use of traffic shaping which may be time- based, user-based, protocol or application- based—and therefore ‘fair’ in the sense that it addresses specific areas of bandwidth Supply consumption.

Packet Cable Multimedia (PCMM) enables bandwidth supply control at the CMTS when applied to specific applications or service tiers. This enables the operators to monetize Quality of Service. In fact, it would be possible to package Quality of Service as a product.

Additionally as noted in Figure 14 it is possible to manage the supply of bandwidth through the use of a virtual or real two-tier backbone. Using this structure content providers and subscribers that pay for Quality of Service are able to access additional bandwidth.

Figure 15. Bandwidth Expansion and Management is Not Cost-Free to the Cable Operator.

Figure 14. Two Tier Backbone Helps to CONCLUSION Manage Supply We outlined a trend in the rise of The Cost of Capacity Management measured congestion and provided an overview of the historical and future Reviewing the capital cost of capacity application demands causing this congestion, management in more detail we see that the and offer suggestions and relationships (a cost of effectively managing capacity is far toolkit) that the cable operator may want to lower than capacity expansion. It follows consider. We note that while HFC networks logically then that options for managing this offer great flexibility in expanding bandwidth, scare resource should be explored before/as not all alternatives are created equal; some are expansion is considered. Table 8 provides easier and more cost effective than others. A examples for consideration. systematic and thoughtful approach to addressing congestion should include $ Per $ Per HP $ Per Data evaluating all possible options—considering Node Per at 2,000 Sub both the subscriber and the content provider in MHz of HP Per (25% pen.) view of both manufactured bandwidth and Capacity Node PCMM xx $166 $1.25 $5 technology bandwidth alternatives. Traffic $20 $0.15 $0.60 Shaping xxi Table 8. The Cost of Technology Bandwidth – Managing a Scarce Resource.

REFERENCES

xi Google press release: “Lights, Camera, i UPC Broadband & C-COR internal reports Action!! - Google Introduces Click-to-Play based on CableEdge® measurements. Video Ads on the Google Content Network” May 23, 2006. ii Bob McIntyre, “The QAM Before The http://www.google.com/intl/en/press/annc/clic Storm”, SCTE Emerging Technologies ktoplay.html Conference, January 2007 xii Wired Magazine Feb 2007 and industry iii John T. Chapman, “Multimedia Traffic sources. Engineering: The Bursty Data Model” , Cisco, January 8, 2002 xiii James Hibberd, “Cable Late-Night Growth Outpacing Prime - Competition From Other iv Ian Fogg, “Home Networks in Europe: Platforms Creates Opportunity for Daypart”, Leverage Strong Growth Potential in TelevisionWeek, August 7, 2006. Chart notes Congested Digital Device Landscape”, Jupiter Nielsen Media Research 2005 Prime-time Research, Jan 2007 households using television rates at 60.2% in 2005 and Late-night households using v Based on historical Kbps per subscriber television levels at 33.9%. growth in Europe. Forecast Kbps numbers http://www.tvweek.com/article.cms?articleId= based on High Speed data applications only 30364 and ignores VOIP and Digital Set-top return path traffic which are assumed to be included xiv Joost is a newly announced television in the remaining 20% port capacity. sharing application that utilizes P2P communication to deliver high-quality vi Andrew Parker, “The True Picture of Peer television across the broadband network. to Peer File Sharing”, CacheLogic, 2004. http://www.cachelogic.com/home/pages/studi xv We assume the average cost to upgrade to es/2004_01.php 1GHz is around $45-$55 per HP. We assume the 1 GHz amplifier gain is higher than those vii A revival in NNTP ‘news’ traffic has been at 870 MHz and permit a direct drop-in at observed recently in Europe primarily as a existing amp locations. mechanism to share multi-part binaries. xvi Fiber Deep calculation based on an viii CacheLogic study: The True Picture of assumed a Labour of 41,000 and materials of Peer to Peer File Sharing. Andrew Parker, 50,000 per 2,000 Home Passed or a total of 2004. 91,000 per 2,000 Home Passed or $46 per http://www.cachelogic.com/home/pages/studi Home Passed. Assumes Greenfield … cost es/2004_01.php would be lower if fiber is already in place. ix YouTube website xvii Assumes the worst case costs for a new http://www.youtube.com/t/about node, including HE electronics, and new fiber if needed, would be $30,000. Numbers may x Based on UPC Broadband Netflow vary depending on specific labor costs, measurements in Europe using YouTube’s AS electronics, and market. number.

xviii Switched digital assumes 80% digital penetration, 2,000 homes passed per service group, 4 QAM channels per service group, 100 switched channels, 10 channels re- encoded. Components include QAM modulators, SDV servers, Re-encoding, Multiplexing, Routing and transport. Assumes 4 x 8MHz of capacity can be released for other services. xix Bonding Channel to Increase Bandwidth based on an assumed $10,000 per 8MHz DOCSIS channel today and assuming $10,000 per 4 x 8MHz DOCSIS channels in the future. We assumed 500 subscribers per downstream or $20 per subscriber. We also assumed $50 per CPE device. xx PacketCable Multimedia: Assuming $5 per subscriber for Turbo Button and Gaming Application. Includes cost of Record Keeping Server. xxi Adding Per-Protocol Traffic Shaping: Assuming 25 CMTS port per shaping device based on 900 Mbps per device and 50Mbps per downstream at 70% utilization or 35Mbps. Each shaping device costs $30,000 / 25 ports = $1,200 per CMTS / 4 nodes per downstream serving segment = $300 per node / 2,000 Homes Passed per node = $0.15 per home passed. We assumed that 25% of the total capacity could be reclaimed though shaping.

THE FUTURE OF TRANSCODING – THE NEED FOR MPEG-2 AND MPEG-4 TO COEXIST John Hartung, Santhana Krishnamachari EGT

Abstract encoders are not required to make use of all the operations available at the decoder. The Virtually all commercial digital video standard was designed with complexity in content is stored and distributed using the mind so that real time encoders could be MPEG-2 encoding standard introduced economically deployed. Early MPEG-2 about 12 years ago. Although this standard encoders were limited in performance due to enabled a large increase in the number of the required computational complexity programs that could be carried in access needed to generate optimal bit streams. In networks, the capacity of those networks has general, this optimality requires a global not kept up with the explosion of content and search over a large number of encoding the increased bandwidth requirement for parameters, and processors capable of this high definition. The introduction of switched computation would have made their cost access networks, both HFC and IP, will help prohibitive. Since the introduction of this alleviate that bottleneck and allow the standard the performance of integrated introduction of new clients that make use of circuits has dramatically increased, and the more efficient MPEG-4 part 10 (a.k.a. algorithms have been developed that achieve H.264) encoding standard. The challenge in near optimal performance at greatly reduced realizing these gains, however, will depend complexity. on transcoding technology that is both cost effective and maintains the quality of the Today’s MPEG-2 encoders achieve near originally distributed program. optimal performance in terms of objective performance measures such as the peak signal to noise ratio (PSNR) as illustrated in THE LIMITS OF MPEG-2 ENCODING Figure 1. This plot shows the distortion, relative to the original content, averaged over The MPEG-2 standard, like all media a representative set of 18 video sequences encoding standards, is defined by the including 24 fps film content, 30 fps encoded bit stream syntax, and the semantics, interlaced content, and clean and noisy or operations, signified at the decoder by sequences. PSNR of about 34 dB results in these bit stream elements. The main semantic high quality encoding that is nearly elements include block based motion indistinguishable from the original content. compensated prediction, quantized transform As seen in the plot, the MPEG-2 encoding based encoding of prediction residuals and algorithm can achieve this result, over a reference blocks, and entropy coding of broad range of sequences, at about 3.5 Mbps encoding parameters. MPEG-2 encoder for standard definition video. operation is not specified by the standard and MPEG-2 Performance

36 35 34 33 32 MPEG-2 31

PSNR (dB) 30 29 28 27 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 bitrate (Mbps)

Figure 1

Objective measures, however, do not tell quantized transform coding of residuals, and the entire story of encoder performance. In entropy coding. However, MPEG-4 addition to the purely objective distortion introduces basic differences in these tools measures, today’s encoders take into account along with additional modes of operation. subjective evaluation through the incorporation of human visual system (HVS) The MPEG-4 motion estimation tools models. These models estimate the masking have been expanded to include additional of distortion by content so that encoding bits block shapes and sizes, and multiple are preferentially spent to reduce the reference frames can be used to predict a distortion that is most visible. Encoders also macroblock. Field and frame prediction can make use of preprocessing to improve their also be varied on a macroblock basis. A new performance on noisy sequences. These intraframe spatial prediction mode has also adaptive filters remove noise from the been introduced that has no correspondence original sequence before encoding, both to to the MPEG-2 coding modes. The DCT restore the original content and to avoid transform used in MPEG-2 has been replaced allocating excessive bits to encode the by a smaller integer transform, and VLC typically high frequency noise. Although the entropy coding has been augmented with an use of HVS models and preprocessing are optional adaptive binary arithmetic entropy not covered in the encoding standards, they coder (CABAC). The MPEG-4 standard also are important to the practical deployment of allows a filter in the encoding loop that helps video encoding. mitigate encoding artifacts, such as blocking at low encoding rates. Although this does not MPEG-4 PERFORMANCE improve the PSNR performance, it results in more acceptable subjective artifacts. Making The MPEG-4 and MPEG-2 encoding full use of these new features enables a 30% algorithms are similar in that they both use to 50% reduction of coding rate for block based motion compensated prediction, equivalent video quality at the expense of a 5-6 times increase in complexity. Figure 2 top box based on a subscriber request. shows the performance of MPEG-4 averaged Switched broadcast enables broadcast over a broad range of sequences and content to occupy a portion of the access encoding rates. The input sequences are full network only when it is requested by one or D1 resolution and the encoder is operating at more set top box clients. In this case, main profile@level 3. capacity is increased beyond broadcasting all channels because only a small number of the It can be seen from the plot that a PSNR programs offered are actually requested of about 34 dB is achieved at an encoding concurrently. rate of about 2.1 Mbps as compared to 3.5 Mbps for MPEG-2, or a rate reduction of Switching both for VOD and broadcast

MPEG-4 Performance

35.00 34.00 33.00 32.00 31.00 30.00 MPEG-4 29.00 PSNR (dB) 28.00 27.00 26.00 25.00 0.5 0.75 1 1.25 1.5 1.75 2 2.25 bitrate (Mbps)

Figure 2 about 40%. can also be used to enable further bandwidth As with MPEG-2, MPEG-4 encoding savings by tailoring the requested content benefits from preprocessing and HVS based on the capabilities of the requesting set modeling and identical techniques can be top box. Because of the large number of applied to improve the subjective quality. deployed MPEG-2 set top boxes, both MPEG-2 and MPEG-4 set top boxes will co- SWITCHED NETWORKS AND exist in cable networks for some time. In MULTIPLE CODECS order to take advantage of the additional bandwidth savings of MPEG-4 set top boxes, Switched networks have been deployed in content needs to be available for either on the cable networks for video on demand (VOD) network when requested. Since switched applications, and are being deployed to applications are aware of the requesting increase the effective capacity for broadcast client, the delivery can be tailored to the applications. Switching for VOD enables capability of that client, e.g. for VOD only content to be switched to an individual set the single requestor need be considered. In the case of switched broadcast all requestors encoding available in MPEG-4, however, a must be considered in order to avoid large set of tools would be restricted from transmitting the same broadcast content in use limiting the ultimate performance. multiple formats. This can be avoided by An alternative approach is to decode the transmitting the content in MPEG-2 format MPEG-2 content and apply the decoded only since most MPEG-4 set top boxes can baseband video directly to an MPEG-4 also decode MPEG-2 content. encoder. This approach does not produce high quality results, as illustrated in Figure 3. In the case of VOD, content can be stored This plot compares the average PSNR of in multiple formats on the server, however, original sources that have been encoded with broadcast applications require transcoding an MPEG-4 encoder, vs. the PSNR of from the predominant MPEG-2 format to decoded MPEG-2 sequences that have been MPEG-4 in real time. encoded with the same MPEG-4 parameters. In this case the MPEG-2 sequences were APPROACHES TO TRANSCODING coded at 4 Mbps and 3 Mbps, and their PSNRs were about equal, or greater, than Several approaches can be taken to that of the MPEG-4 encoding. This result transcode from MPEG-2 to MPEG-4. The uses the same set of sequences used in lowest complexity approach involves Figures 1 and 2. As the plot shows, the mapping the MPEG-2 encoding parameters PSNR degrades up to 1 dB in the into MPEG-4 equivalent representations. decode/encode case when the MPEG-2 was

MPEG-4 Encode v.s. MPEG-2 Decode/MPEG-4 Encode

35.00 34.00 33.00 32.00 31.00 MPEG-4 Encode 30.00 Transcode from 4 29.00 Transcode from 3 PSNR (dB) PSNR 28.00 27.00 26.00 25.00 0.5 1 1.5 2 bitrate (Mbps)

Figure 3

This is similar to the encoded domain rate coded at 4 Mbps. One of the main reasons for shaping that is used for statistically the degradation is due to the fact that the multiplexing MPEG-2 streams. Encoding frame type is not maintained between the gains can be obtained through the intraframe MPEG-2 and MPEG-4 encodings. The B prediction modes and improved entropy frames are typically encoded at lower rate, and PSNR, than I and P frames in the larger gains are possible. Passing additional original MPEG-2 encoding. Without parameters allows for further improvements knowledge of the frame type, the MPEG-4 in transcoding performance, and a reduction encoder can re-encode the B frames as I or P in complexity. An example of this is the use frames and subsequently use these as of bit allocation in the MPEG-2 encoding as reference frames for prediction. This results a complexity estimator that can be used for in lower PSNR in the predicted frames and rate allocation in the transcoded sequence. propagation of this distortion. The results This is similar to two-pass encoding shown at 4 Mbps are for high quality MPEG- algorithms, however, the necessary 2 encoding that maintains good quality for I, information already exists in the MPEG-2 B, and P frames. Lower MPEG2 rates, and/or bitstream. This technique improves the lower quality encoders produce a larger transcoding quality without the complexity variation in the quality of the different frame of a two-pass MPEG-4 implementation. types resulting in poorer transcoding results as shown in the 3 Mbps result in Figure 3. As Overall, high quality transcoding and illustrated in the plot, the resulting PSNR of reduced complexity is achieved through full the transcoded sequence degrades further at decode/encode with reuse of encoding the lower MPEG-2 rates. parameters. This argues for a tightly coupled system that receives MPEG-2 programs, A final method for transcoding also either in SPTS or MPTS, and converts performs decoding to baseband video and re- directly into MPEG-4 transport streams. encoding using an MPEG-4 encoder, however, the MPEG-4 encoder makes use of A final consideration in transcoding is the the MPEG-2 encoding parameters. One mitigation of source noise and coding example of this is to maintain the frame artifacts in the original MPEG-2 encoding. coding type to avoid the degradation This can be accomplished by filtering the described in the previous method. Referring decoded MPEG-2 sequence, either in the to the results in Figure 3 again, transform domain, or in the reconstructed approximately .2 dB improvement can be baseband for encode/decode transcoders.

Figure 4 Figure 5 gained when transcoding from 4 Mbps High quality MPEG-2 encoders typically MPEG-2 to 2 Mbps MPEG-4. For lower rate apply sophisticated prefilters that remove and/or lower quality MPEG-2 encoding even noise, however, noise filtering improves transcoding when this is not the case. A Increasing amounts of high definition content second source of noise can be introduced by further add to the need for improved the MPEG-2 encoder. This structured noise efficiency. Switched services will help can be effectively estimated and adaptively provide this additional bandwidth and enable removed to improve the subjective quality of a transition to the more efficient MPEG-4 the transcoded sequence. Figures 4 and 5 encoding standard, however, this transition show the corresponding frame in a will require the coexistence of both MPEG-2 transcoded MPEG-2 sequence. Both frames and MPEG-4 services. This coexistence will use the same original and transcoded be enabled by integrated transcoders that parameters, however, Figure 5 demonstrates provide a high quality, low complexity the improvements gained through post- solution. filtering of the MPEG-2 sequence. CONTACTS CONCLUSION John Hartung The need for bandwidth efficiency in [email protected] cable plants continues to be driven by the Santhana Krishnamachari increase in service and content offerings. [email protected]

THE IMPACT OF GOING FIBER SPEEDS OVER COAX FEEDS Alon Bernstein, Cisco Systems, Inc. Joe Godas, Cablevision

Abstract Most of the issues described here are not unique to cable. Any service provider Data rates over cable networks are network that will serve 100mbps+ speeds constantly increasing due to higher will have to deal with them. However, bandwidth demands from consumers as well DOCSIS 3.0 is the technology that would as competitive pressure. With DOCSIS 3.0 allow, for the first time, residential the cable plant can offer speeds that rival costumers to have access at these rates and those offered by PON (Passive Optical so it will be the first time cable service Networks) technologies and meet consumer providers manage such a high speed demands. network.

Building an end-to-end solution that will IMPACT OF HIGHER SERVICE support these data rates is not only about CONTRACTS ON AGGREGATE adding faster network components and TRAFFIC updating the user service agreements. This paper will explore the specific impact of Even before DOCSIS 3.0, in order to supporting data rates in the order of 100 compete with telcos, cable Multi-Service Mbps and above over the cable network. Operators (MSOs) have been steadily increasing their traffic contracts from 2.5mbps down/500k up to 30mbps down/5 OVERVIEW mbps up. Looking at the lessons learned from past increases will help to infer what With DOCSIS 3.0, the typical traffic will happen with future increases. profile for a cable subscriber is going to increase from the 10mbps-30mbps range to While one might expect that the the 50mbps-100mbps range. These traffic aggregate trend would linearly track the rates compete well against end-to-end service contract increase, or in other words, optical technologies. that a doubling of a service contract will result in a doubling of the aggregate traffic, In theory it takes a DOCSIS 3.0 system it is evident this is not the case. and a change in the modem configuration file to get to the above rates. In practice, the The following graphs represent aggregate added throughput has implications on the traffic as measured at the internet border in upper layer application behavior, network the Cablevision network at different periods design and equipment design. This paper over the years. will list what these implications are and how they can be addressed in order to unleash the full potential of DOCSIS 3.0.

Figure 1: 1Q2004

Figure 1 depicts the aggregate traffic the system. The modem contracts haven’t during 2004Q1 when an individual contract changed but downstream consumption has was 10mbps down, 1mbps up. As can be almost doubled due to a 45% increase in seen, the traffic peaks are around 12Gbps subscriber acquisition (see Figure 5). Now and the lows are in the 4Gbps range. The the peaks are in the 22Gbps range, and the gap between the highs and lows is roughly lows at 7Gbps. The spread has increased to 8Gbps. 15Gbps, much more then the linear increase 18 months later we see traffic demands in subscriber bandwidth. ramp up quickly as subscribers are added to

Figure 2: 3Q2005 spread is 38Gbps, and keeps growing In 2Q2006 (Figure 3) -The traffic much faster then expected. In this time contracts for the entire subscriber base period, a 19% increase in subscriber were increased to 15mbps down and acquisition has produced a whopping 2mbps up. The peaks rose to about 100% increase in traffic consumption. 40Gbps, and the dips to 12Gbps. The This surge in demand is attributed to a 50% increase in the modem contract 15Mbps downstream contract). burst capability (going from a 10Mbps to

Figure 3: 2Q2006

1Q2007 is especially interesting. The Curiously enough, while the peaks have first thing to note is that it is barely 6 increased by 10Gbps, the lows have months since 2Q2006 and corresponds hardly moved from their 2Q2006 range to a subscriber increase of roughly 9%. (about 12Gbps).

Figure 4: 1Q2007

Several observations can be drawn highly interactive applications (file from these set of traffic trends: downloading or real-time streaming) are the dominant behavior during peak time. 1. Traffic rates are increasing rapidly, Higher burstable contracts appear to and non-linearly; have the effect of creating higher peak bursts, but not higher sustained use (as 2. Across all graphs, the peaks and the noted by stagnating dwell/gap growth). lows occur at about the same time, so they most likely correspond to Another important take away from interactive use of the internet around this graph is that network capacity evening time; planning techniques will need to change. The rules previously used for lower 3. If the majority of traffic was peer-to- bandwidth contracts cannot be extended peer (P2P) then the lows would have to higher modem contract rates because been much more pronounced as people of the increased burst capability (thus tend to leave their P2P clients on higher aggregate peak rate) of the traffic through the night. Either people shut flows. down their P2P when not active, or some

%%(Kbps) SUBS Delta -Subs Sub incr Aggregate BPS Delta - BPS BPS i ncr BPS/Sub 2004 Jan 1,082,662 - - 12,000,000,000 - - 11.08 2005 August 1,569,475 486,813 45% 22,000,000,000 10,000,000,000 83.33% 14.02 2006 June 1,866,716 297,241 19% 44,000,000,000 22,000,000,000 100.00% 23.57 2007 January 2,039,259 172,543 9% 56,000,000,000 12,000,000,000 27.27% 27.46

Figure 5: Usage/Growth stats

TCP/IP PERFORMACE IN HIGH problem of having a limited “ACK rate”. SPEED NETWORKS See RFC3449 (TCP Performance Implications of Network Path DOCSIS 2.0 Transmission Control Asymmetry) for more detail on TCP Protocol (TCP) downstream throughput performance limits due to asymmetrical is gated by the number of links. Acknowledgment (ACK) packets that can be sent on the upstream. This is DOCSIS 3.0 allows multiple because TCP tends to self-clock and the outstanding requests and so the rate of the segments sent on the bottleneck imposed by ACK rates is downstream depends on the rate of greatly reduced. The TCP receive ACKs sent on the upstream. In other window size becomes the new words, if a cable modem can send up to bottleneck. The following text explains 200 ACKs per second, the downstream the TCP receive window issue at a high rate can not exceed “ACK per second X level: segment size” Bytes per second, or in a typical case 200*1460*8 = 2.33Mbps. TCP defines a “window” as the DOCSIS Upstream Concatenation can amount of bytes that can be sent help, but it does not solve the basic unacknowledged to a network. They don’t need to be acknowledged (yet) To make things worse, the default because they count as in-transit, meaning window size in Windows XP is set to that due to the various delays in the 16K, which will limit the maximal network they either did not have a throughput in the example about to 52/4 chance to be received by the end point, = 13Mbps (even below DOCSIS 2.0 or the ACK did not have enough time to capabilities). travel back. In fact, TCP attempts to keep this window at a maximum because This issue with TCP has been that way it “fills the pipe” between the recognized a long time ago, and in 1992 sender and receiver so that there are the IETF published RFC1323 that always bytes in-transmit. defines TCP/IP extensions to allow larger windows sizes. The original TCP definition sets the maximum number of bytes that can be To demonstrate that this in not a cable in-transit through the network to specific problem, even on the web pages 64Kbyte because of the 16bit encoding of FiOS (Fiber to the home service from of the window size in the TCP header. If Verizon), it is recommended to run an we assume a very simplistic model ActiveX component (Figure 6) that sets where a network introduces a 100ms the proper entries in the windows roundtrip delay (possible on the Internet) registry in order to take advantage of the and that a single ACK clears a whole RFC1323 support in Windows. 64Kbyte window, then the maximal best Note that they are other ways to increase case throughput of a TCP session can TCP throughput and any “TCP speedup” not exceed 64Kbyte * 100ms * 8 = utility would try setting the same 52Mbps, which is less then the parameters that are set by Verizon. capability of a DOCSIS 3.0 modem – and is a theoretical upper bound. In a real system it’s likely to be even slower.

Figure 6: Verizon TCP speed up page

The following parameters are set: is limited to 54Mbps, it is possible to congest the home network with data. 1. TCP 1323 Extensions - This parameter enables enhancements To allow for proper QoS handling in to the TCP/IP protocol that the event of a home network congestion provide improved performance DOCSIS 3.0 has the option of over high speed connections; embedding priority bits in the 1-Byte, 3- Byte or 5-Byte DOCSIS headers. The 2. TCP Receive Window - This priority bits are taken into consideration parameter specifies the number when queuing packets after the packets of bytes a sender (the source you have been processed on the RF side and are downloading from) may are on the way to the home network. transmit without receiving an Once they are queued to the home acknowledgment. Modifying it network, than in the event of congestion determines the maximum size the lower priority packets would be offered by the system (appears to dropped first. be about 400K); The priority bits are arranged based on the Ethernet 801.D priority bits 3. MTU (Maximum Transmission definition as depicted in Table 1: Units) - The MTU defines the largest single unit of data that can Number of CM output be transmitted over your queues connection. The FiOS network 2 3 4 5 6 7 8 requires an MTU of 1492 bytes. 0 0 0 0 0 0 0 0 All utilities are available on the web (Default) under the general categories of “TCP 1 0 0 0 0 0 0 1 optimizer”, “TCP speed” etc, change 2 0 0 1 1 1 1 2 these parameters. In a cable environment, and especially with 3 0 0 1 1 2 2 3 DOCSIS 3.0, a user should update the 4 1 1 2 2 3 3 4 operating system TCP defaults to work with TCP 1323 extensions, so that larger Traffic Priority 5 1 1 2 3 4 4 5 window sizes can be define. 6 1 2 3 4 5 5 6

IMPACT ON THE HOME NETWORK 7 1 2 3 4 5 6 7

Table 1 : DOCSIS 3.0 priority encoding The bottleneck link at the home used to be the internet connection. It was fair This table allows vendors to build to assume that whatever the internet devices with varying degrees of delivered, the home network can complexity in terms of how many output consume. With DOCSIS 3.0 this might queues they have on their home network no longer be the case. For example, if a interfaces. The table defines which user has a contract for a burst rate of priority encoding goes to each queue. 100Mbps, but the home wireless router Natually, if there are 8 queues then each lifting” per packet/byte is increasing. A gets a dedicated priority queue. sample of application that require per byte operation are: This feature can work in tandem with uPNP and Ethernet type flow control to 1. Deep Packet Inspection: For preserve end-to-end QoS all the way to reasons that relate to filtering certain the home device. network traffic type, and to protecting the network against Denial of Service HIGH SPEED AND POWER Attacks, the network has to look beyond CONSUMPTION the L2/L3 portion of a packet in order to figure out what a packet is, not just Another side impact of higher traffic where it goes to. rates is the increased power consumption 2. Video Streams manipulation: by networking equipment. As with other High touch video processing, such as issues raised in this article, this is not a transcoding, trans-rating and ad-insertion cable specific issue. It’s not even a require extensive byte manipulation. communication industry specific issue. It 3. Encryption: While not new to is a result of the current CMOS cable networks, encryption is a byte-by- technologies, used in most of today’s byte operation on a whole packet. The ASICs, reaching their physical faster the data rates are, the faster the limitations. encryption chips have to run and the more power they need. For example, modern oxides (the insulation layer in CMOS) are at 10-12 Hopefully, technology innovations angstrom, that’s 5 atoms thick! In will help reduce the power requirment as addition to the fact its hard to go much much as possible. In addition,new lower then 5 atoms, there is already a network architectures, such as M-CMTS, scaling problem, since reducing the allows for distribution of function density by half would require placing a (L1/L2/L3) which is also a distribution 2.5 atom thick oxide – clearly not an of power. For example, if the M-CMTS option…There is also a limit to the packet shelf is located outside the hub, it minimal voltage that a CMOS circuit can reduces the power demands on the hub. operate in. As a result, there a limit to how much power can be reduced – a limit that impacts CPUs, mobile devices, DUAL TOKEN BUCKET data centers and telecom. A single token bucket definition (as In the past routers/switches had to the one used in DOCSIS 1.1/2.0) places look only at the packet headers restriction on the peak rate and burst size (Ethernet/VLAN for switches, IP for of a flow. The problem with a single routers) in order to forward a packet to token bucket definition is that for the its proper destination. As data rates went duration of a burst the flow is not limited higher, the faster the forwarding rates to the peak rate, instead, the burst rate had to go and the more power was can be as high as the total capacity of the needed to perform it. In addition to this link. As long as the burst size is minimal basic forwarding the amount of “heavy (in the range of a couple of Ethernet frames) this is not a significant issue, but applications that might use the for those customers who define large bandwidth provided by DOCSIS 3.0 and burst sizes the burst size could become their impact on the network. In one an issue with assuring fairness across a sentence, the one common theme is that large number of flows. It can further be video is the new “killer app”that will exacerbated because of the data rates drive bandwidth demands. supported in DOCSIS 3.0. Today, a majority of internet traffic is The way to address extended bursts related to file sharing. In fact, one may with the DOCSIS 3.0 toolkit is the newly argue that a good percentage of file defined “maximum sustained traffic sharing traffic is video content and rate” and “maximum downstream traffic therefore is a form of video on demand rate”. These two parameters refer to how distribution. In that sense we are already fast the traffic can flow during the traffic experiencing a bandwidth explosion that vs. how fast it can flow once the burst is is video driven. The next step to file exhausted. For example, a traffic sharing is to stream the video content contract can be: and play it while it’s being downloaded, ƒ Maximum sustained traffic rate is and this is already being implemented by 10mbps joost (www.joost.com). The “joost” type of video streaming has a couple of ƒ Maximum downstream traffic fundamental impacts on network traffic rate is 30mbps patterns.

ƒ Burst size is 2Mbytes 1. Since every home is a media source, as well as media consumer, the upstream For this contract, the user can send up to bandwidth is driven higher. 2Mbytes of data at 30mbps, but if the user sends more then 2Mbytes the 2. Joost, just like a file sharing CMTS would reduce the rate to 10mbps. application, opens up multiple parallel TCP sessions. That means that it will try to squeeze as much bandwidth from the TRAFFIC PATTERNS IN A HIGH existing traffic contract without trying to THROUGHPUT NETWORK be “fair”.

With the bandwidth that DOCSIS 3.0 While joost and the multitude of provides it is very likely that a new crop joost-like applications that are sure to of applications will show up and put to follow, represent “over the top” video use the added capacity. These new (meaning video sources that are not applications will have new traffic originated by the cable MSO), a cable patterns, and therefore the network MSO IPTV delivery will have its own planning can not be derived from the impact of bandwidth usage and traffic existing traffic patterns. The first section patterns. Although its not clear at the of this paper already demonstrates how moment if IPTV will be a significant unexpected the aggregate traffic can be bandwidth drive in the cable MSO when individual contracts are increased. world, its worth exploring the way it is This section will discuss some of the different then over the top delivery. MSO generated IPTV streams are likely to be carried over RTP (as opposed to TCP). While TCP is a closed loop protocol, RTP is open loop, meaning that the RTP source will not slow down or try to be network friendly if the network is congested. In that case, one might ask why use RTP instead of TCP? The reason is that RTP has a much better story when it comes to delivering real- time content and that is one way to differentiate MSO content from over- the-top content. But because open-loop delivery can not recover as gracefully as TCP, the network has to be over provisioned by much more then with TCP flows. In fact, if IPTV catches on, there could be a major shift from a world where most traffic is TCP to a world where most traffic is RTP: instead of flows that regulate themselves (TCP) a network that has to be over-provisioned to accommodate non-flexible traffic flows (RTP). Also take into account the fact that a video flow is fairly long lived, and the end result is a network that future networks can’t be oversubscribed by the amount used today. Note that this conclusion may be somewhat counter- intuitive since higher bandwidth, in principal, should lead to better statistical multiplexing and higher oversubscription ratios. If the internet traffic remined as it is today, this may have been the case, but it is less likely when considering future applications.

-Alon Bernstein, [email protected] -Joe Godas, [email protected]

THE UNIQUE CHALLENGES FACED BY CABLE SYSTEMS SERVING SMALLER MARKETS AS THEY EXPAND TO MEET CONDITIONAL ACCESS AND OCAP REQUIREMENTS Gary Traver, James Capps Comcast Media Center

Abstract INTRODUCTION

Separable security systems represent a U.S. cable systems serving exurban areas fundamental component of a long-term of the U.S, the first markets to embrace migration towards an open standards / open cable television’s “community antenna” architecture world. OCAP, the OpenCable technology, face the greatest challenge in Application Platform, which was developed creating next-generation content distribution for the industry by CableLabsi, is designed networks. As cable nears its 60th to drive the future of interchangeable anniversary in many of these communities, devices toward this vision for an open- these cable system operators are tasked with architecture environment. meeting new conditional access requirements that potentially thwart their The operational rigor associated with ability to offer advanced digital video managing devices in this interchangeable services and threaten their survival. In world is complex. As both separable addition, they must address expensive security and OCAP evolve, the number of regulatory requirements at the same time consumer devices able to make use of this that their customers have more choices for open network will grow. the providers of those services than any other time in historyiii. Cable systems serving smaller markets must address the expensive regulatory While separable security systems requirement for separable security and the represent a fundamental component of a migration toward OCAP at the same time long-term migration towards an open that their customers have more choices for standards / open architecture world, the the providers of advanced video services entire cable industry must respond by than any other time in historyii. complying with a FCC mandate for using separable security to serve all new video Because they lack the economies of scale customers in the short-term by July 7, 2007. afforded by cable systems serving larger For the purposes of this paper, we are markets, cable operators serving smaller assuming that separable security communities need solutions that build upon methodology will evolve over the next their existing architecture. In addition, they several years into a fully software- can meet or exceed the economies of a downloadable security environment. large-market solution by pooling their resources into a centralized service bureau. In addition to managing the conditional However, a one-size-fits-all solution is access and security processes, another key counter-productive to a competitive element of enabling the next-generation marketplace. To be effective, the centralized content distribution architecture is its ability services bureau will also need to address the to support the open standards environment unique requirements of each cable system.

envisioned by OCAP, the OpenCable use of consumer-owned devices on the Application Platform. OCAP, which was network have very similar requirements and developed for the industry by CableLabsiv, thus very similar cost issues for cable is designed to drive the future of operators serving smaller markets. interchangeable devices (i.e., STBs, PCs, TVs, etc.) that can easily connect to a cable Finding an economically viable solution operator’s network and download that for providing these services requires some operator’s user interface and the key system key changes to the smaller markets’ systems information that is necessary for the device and architecture. To provide the same scale to navigate within the operator’s network. economics as larger systems, the conditional This future downloadable, interchangeable access solutions must be able to pool world creates a host of new opportunities. subscriber bases together from many small, However, the ability to connect any device unaffiliated cable systems. It must do so in that is both security-compliant and OCAP- a manner that maintains the individual compliant also creates a level of complexity configurations and attributes of each small well beyond the scope of anything that is in system and maintains the security and place today. integrity of each system’s data. The most viable approach to meet these criteria is to Cable operators serving smaller markets create a centralized service bureau type of must comply with security requirements operation. without the same economies of scale as regional cable systems, which serve an The operational rigor associated with urban/suburban cluster of customers from managing devices in this interchangeable one headend. For example, the minimum world is complex. As both separable cost of the headend equipment and software security and OCAP evolve, the number of required to meet federal requirements for consumer devices able to make use of this separable security are fixed. Therefore, the open network will grow. Attention to investment required to implement those maintaining network compatibility must be solutions can be several magnitudes higher focused in two directions. The focus must on a cost per subscriber basis for a small first be technology forward, to make sure cable operator than they are for a larger that the system is capable of supporting new cable operation. If the minimum system consumer devices as they continue to grow configuration is sized to meet a 25,000 and evolve. It must also be technology subscriber base, operators with a 5,000 backward, to ensure the legacy equipment customer base must pay roughly 5 times the in the cable systems continues to receive the cost per customer, and systems in the 1,000 appropriate support. Cable operators customer base range must pay 25 times the serving smaller markets must have solutions cost per subscriber in order to offer the same that allow them to build upon their existing mandated solutions. architecture. The economics of their local businesses do not allow them to engage in a Accordingly, the same economic issue total overhaul of their current content that applies to separable security also applies distribution infrastructure. to the larger implementation of OCAP in these exurban markets. The systems The country’s independent cable necessary to detect, manage and deliver the operator community must have an other critical layers of software to enable the affordable solution that complies with

federal regulations and additional choices that consumers will have to requirements regarding conditional access experience the array of services provided by security systems by July of this year. While their cable operators. Further, as this many of these systems have expanded their environment begins to proliferate across the businesses to offer high speed data and entire marketplace, all operators will voice services, regulatory requirements pose eventually be expected by their customers to a real threat to their continued existence. In provide support for consumer-supplied fact, as the American Cable Association devices and the type of portability that they observed: “...many of the small businesses provide. Customers who already have their that provide video and broadband services in own downloadable devices will exacerbate rural America will cease to exist and the this demand even further when they move digital divide will actually growv.” into another cable system’s franchise area.

As this paper will demonstrate, a solution However, providing the systems and that is focused on the unique needs of the infrastructure required to provision services smaller market in the form of a centralized to a wide array of devices in this future multi-operator services bureau can allow downloadable environment is not an easy or many cable systems to not only overcome inexpensive task. The complexity of the these near-term challenges in conditional equipment and the systems required to access but also provide a long-term strategy support an open environment increases for OCAP success. This robust services significantly over current operational platform must address many of the requirements. Without economically and requirements for migrating to an all-open operationally viable alternatives, this standards downloadable environment. By migration could result in cable system pooling these markets into a national operators struggling to provide even the platform, it will also allow them to offer the most basic services under this next- same access and advantages as other larger generation architecture. cable organizations, enabling smaller system operators to expand the lineup of advanced From the author’s perspective, support digital services that they can offer to their for separable security and for OCAP are customers. extensions of the same core mission. They both allow for portability and they both THE CHALLENGES AND require the consumer device to be connected OPPORTUNITIES OF NEXT- to the cable operators’ network in order to GENERATION CONDITIONAL ACCESS get the necessary software and information required to properly receive the services While there are many challenges provided by the operator. Most importantly, associated with creating secure and robust the consumer devices must depend on the systems that meet the federal mandate for equipment and systems from both of them to separable security, the intent of the mandate function properly. Therefore, from the is consistent with the cable system standpoint of creating an enabling operators’ need to provide a service that architecture to provision the devices, it responds to their customer’s needs and makes sense to combine their functionality demands. The migration towards separable into a single service entity. security and OCAP-compliant consumer electronic devices increases the number of

ADVANCING TECHNOLOGIES FOR (DCAS). DCAS allows the cable operator SET-TOP OPERATIONS to download its conditional access system of choice to devices connected to its cable A. CableCARD Implications network. It is designed to operate with cable set-top boxes, integrated retail DTVs and Section 629 of the 1996 telecom law other devices that include a secure DCAS resulted in the creation of CableCARD by microprocessor chip. The DCAS protocol is pressing for separable securityvi. available to any consumer electronics Responding to this drive, the OpenCable manufacturer enabling a large market of standards were created and specified the devices to be available for consumers. CableCARD as a removable security device as follows: A relevant advantage to DCAS is responsiveness in that it is able to address • A Host device (set-top box or integrated security breaches more quickly and television) that provides generic cable efficiently. In March of 2007, hackers tuning and decoding capabilities that are cracked encryption codes used by Swiss portable across all cable networks. cable operator Cable COM and digital television technology group Kudelski SA, • A removable CableCARD security and released them on the Internetvii. Web module that separates retail delivered savvy pirates could transfer the published set-top box (Host) from proprietary codes to a digital TV decoder called a conditional access systems and network Dreambox DN 500c, and freely access messaging. subscription video on the cable system’s lineup, according to press reports. A DCAS • An interface between the Host and model allows such breaches to be addressed CableCARD module, that allows for immediately and remotely, requiring no new Hosts and CableCARD modules from hardware or truck-rolls. different vendors to interoperate. C. Expanding the Set-top Functions - Enabling the CableCARD functionality OCAP will require interaction with various downstream systems. Beyond the CableLabs originally developed the additional capital outlay required for the OpenCable family of standards to provide a CableCARD itself, the smaller operators common basis for digital cable TV systems will now need to ensure that their servers are in the U.SSupport for devices addressed via capable of interacting with the set-top boxes a new standard, the OpenCable Application to enable EPG and authorizations. Platform or OCAP. OCAP results in a stack of software residing between applications B. A Future Alternative to CableCARD – and the operating system within a consumer DCAS electronics device such as a set-top box or OCAP-compliant TV set. OCAP devices can A cost-effective, cable company agnostic have new information or applications loaded and conditional access system agnostic on them, often taking advantage of new two- alternative to the CableCARD is the way capabilities. Downloadable Conditional Access System

Figure 1 - High Level Diagram of OCAP Implementation

Additionally, the OpenCable downloads from both the systems managing environment allows multiple vendors to play the security and the systems managing in the headend space, providing functions OCAP functions. It must then continue to and features previously provided by one receive a series of data streams which device or vendor. As with many provide necessary information and software technological advances, increased choice required to sustain functionality. Therefore, bears increased complexity. Figure 1 from the standpoint of creating an enabling highlights the new approach to set-top box architecture to provision the devices, it development under an OCAP system. makes sense to combine their functionality Various providers, vendors and systems now into a single service entity. play under a defined specification to enable previously “built-in” functions. A. Separable Security Deployment

UNIFYING THE SECURITY AND OCAP The first step an operator will need to IMPLEMENTATION undertake is the deployment of an advanced security capability. The implementation of Separable security and OCAP are either separable security solution, DCAS or currently viewed as independent issues and CableCARD, is not a trivial task. Figure 2 projects by the industry. However, when shows the messages required to drive the examined from the deployment and CableCARD. While many of these sustaining operations perspective, they can messages and streams are already a function be viewed as extensions of one another. For of existing systems, the OCAP compliant an advanced security OCAP compliant infrastructure may result in each message device to connect into a cable system’s generator being independent. network, it must first receive a series of

Figure 2 - Device Message Requirements

Independent servers providing EPG connected to the operator’s network, the (Electronic Program Guide) listings data, consumer device must communicate with an and SI (System Information i.e., channel authentication proxy server located in the maps) are now necessary to ensure headend to initialize use of a system. This CableCARDs are initialized and operate in DCAS authentication proxy server sends the the consumer’s home. CAS download key to the set-top. Then the encrypted CAS image is sent to load the set- Figure 3 is the diagram provided by the top box with appropriate encryption FCC in its “Report of the National Cable & software. Telecommunications Association on Downloadable Security”. This diagram highlights the steps required to initialize the DCAS device. As can be seen, once inserted into the consumer device and

Figure 3 - DCAS process as envisioned by the FCCviii

Once the security system has linked with, of software. In some cases, the consumer and authenticated, the consumer device, it device must acquire the OCAP software can then download the series of entitlements stack, a middleware software component and other secure information necessary to developed to enable multiple applications to work in conjunction with the configuration interact together. Applications, like a information included in the services. programming guide or an on-demand ordering system, sit on top of the Therefore, the first step in providing middleware. Operating systems (OS) are advanced conditional access requires that below it. The job of the middleware is to several new applications and servers be translate what lies at the root level for what installed and operated by the operator. To sits above it. This, for example, allows an simply initialize and begin communication interactive trigger from a programmer to with the set-top box in a separable security function without needing to know what environment, the operator must now be version of OS is being used in a particular prepared for an increase in equipment and model of HDTV or set-top box. expertise. Once the stack is available on the device B. Basic Set-top Functions under OCAP – and fully operational, the key software Middleware and Application Installation applications that will remain resident in memory in the consumer device will need to Following the initialization of the be loaded. Key functions such as the main security elements, the consumer devices user interface, including the EPG, and other must be given “personality data” in the form basic functions to the network, like the

applications required to receive and decipher channel maps. Figure 13 shows in additional detail the OCAP stack.

Figure 4 - OCAP Stack with Applications

Again, additional hardware will be one or more servers to provide these types of required in the headend to load the applications to the consumer devices. middleware or applications. The software will be deployed using carousel file servers. Comparing this implementation to Two methods are available for delivering today’s basic cable implementation shows a these applications into the consumer device glimpse of the increased complications that via the carousel file server. The first method come with the new age of cable television. uses the out of band data channel. The The exurban operator will be challenged like second uses the DOCSIS Set-top Gateway never before. Even the larger system (DSG). operators are concerned about the new challenges that lie ahead. In a recent article Once the middleware is installed on the in Communications Technology, Chris set-top box, additional applications that are Bowick, Cox's CTO, was quoted as saying: stored in memory in the consumer device “… (OCAP) [is] a very complicated topic, can be installed. Examples of such and it's not just new set-top boxes and new applications are the main user interface or software on set-top boxes …From a systems the EPG. The consumer device also has the perspective, you have the addressable capacity to download other applications that controller and the DSG that will be used in will reside in memory during the time that OCAP deployments for the two-way they are in use by the consumer. An communications instead of the proprietary example of this type of application is S-A and Motorola approaches we have interactive games. The system will require todayix.”

Implementation of either the communications with local and regional downloadable security system or OCAP security systems. This approach is beyond relies heavily on components not in use the technical and financial limits of many today by a cable operator. While every rural operators. effort will no doubt be made by the numerous vendors involved in developing An OpenCable implementation, as these systems to ensure smooth depicted in Figure 5, is representative of interoperability, the complexity of the initial what larger operators will most likely deployment and long term operation will tax deploy. In addition to providing the even the largest of system operators. Smaller fundamental services such as encryption and market operators now managing their EPG, these cable systems will expand their headends as “lights-out facilities”, requiring competitive service mix by enabling little or no daily management or support, customers with VOD, DVR and OCAP. will find that the basic activities of simply However, Figure 5 also shows the number of adding authorizations or new set-top boxes headend servers and services growing will require complex interactive substantially.

Encoder Video Feeds

Video QAM

Key Management

Conditional Access Settop Box OOB Demod / Modulator HFC Television

ISP

` Cable Modem PC

EPG Server

Carousel File Server OCAP Application Server

DVR Management Server

VOD Systems Figure 5 - Complex OCAP Implementation

DAY TO DAY OPERATIONS - rights needed to access the content. Through ENTITLEMENT AUTHORIZATION the CA system in use (DAC, NAS, DNCS) the ECM ciphers with the broadcast key The small operator’s task list has grown according to the profile and generates a under the new FCC and OCAP guidelines. control word (i.e., the content encryption key). The controller sends the ECM and • First and foremost, the operators must starts streaming the content ciphered with maintain control and function of the control word. existing set-top box population. This value of this investment cannot be Following that communication, the understated. CableCard deciphers the ECM (using the broadcast key) and checks the required • Second, the operator must respond to rights relatively to the list of rights stored in FCC-mandated changes surrounding the card. If the rights are correct, the card security. Changes in headend gear will returns the CW to the decoder that will use it be necessary in order to deploy the new to decipher the received content. When security system(s). subscriber rights have to be updated, an EMM is sent to the user’s CableCard. It is • Third, the operator will need to support ciphered with the broadcast key and contains OCAP product offerings. Advanced the updating information (new rights, loss of services, which will drive revenue, will rights, increase credit limit, etc.) and the require the support of OCAP servers in Subscriber_Id. the headend. B. OCAP Application Authorization and • Fourth, the operator must provide Operations continuous updates to support the devices and functions that will be The previous section addresses the steps offered as the market continues to to authorize a service. This process is evolve. largely handled by the CA system with interfaces to new external devices. The • Finally, the operator needs to ensure authorization of applications, and the continued operation of this system to submission of necessary data to those authorize and manage these services and applications, is a much larger and more functions. complicated endeavor. Another look at Figure 5 shows the increased number of It is this aspect of continued operation servers providing services to the consumer that will be most taxing to an operator. New devices. Previous sections discuss the fact processes and procedures required to add that the applications must be loaded on top set-tops, replace devices and ensure of the OCAP stack. However, the appropriate billing will overwhelm the entitlement of the cable customer to receive lightly staffed smaller cable systems. this application and the associated data streams must also be managed. Interfaces A. Set-top Authorization with billing systems and entitlement servers will be required to enable customers to As with a non-CableCARD set-top box, receive the VOD client and EPG in the service provider defines the types of accordance with appropriate authorizations.

U.S.x serve fewer than 25,000 customers A centralized authorization management and a significant number of them serve system will assist operators greatly. fewer than 5,000 customers. The American However, because each application is likely Cable Association (ACA) represents 930 to be developed independently, a seamless independent cable systems operators that and unified means by which each carousel serve 8 million U.S. cable households via file server can be managed and controlled is 5,000 systems. This translates to an average not likely to exist for several OCAP of 1,600 customers per headend and ACA generations. indicates that more than 1,000 of these systems serve fewer than 1,000 ECONOMIES OF SCALE FOR THE subscribersxi. Many of the ACA’s members SMALLER SYSTEMS OPERATOR also participate in the National Cable Television Co-Operative (NCTC). NCTC A. Centralized Service Bureau represents 1,100 independent cable owners of 5,500 individual systems serving more As previously mentioned, the economics than 10 million subscribers, which also for deploying next-generation content illustrates the number of cable systems distribution systems are very different for serving smaller, exurban communities cable system operators serving exurban across the country. markets within the U.S. In fact, the vast majority of the 7,090 cable systems in the

Figure 6 - Broad Deployment via Centralized Service Bureau

When we are examining how these for separable security and OCAP, it is markets will comply with the requirement important to note the additional limitations

of their current system architecture. In fact, the most recent wave of upgrades in these Regardless, each cable system needs to smaller markets was using 450 MHZ plant be treated as an individual entity. and equipment that came onto the market The number of and type of services when larger systems upgraded to 750 deployed and the timing of each MHzxii. Many of these small-market implementation will be unique for every systems deployed digital video service using market. The support structure of the service the first generation of digital set tops boxes. bureau should be supportive of each cable system’s need to respond to local market The service bureau provides an demands, as operators move to offer choices economically viable solution to cable that match the competitive alternatives. operators serving smaller markets by pooling many small systems to achieve Unilateral services such as unified server economies of scale. This creates a single management, application certification and large installation of equipment and testing all ensure a quality offering to each applications that mirrors or exceeds the headend. It is important that each economies of scale found in a large market. configuration be validated and tested in a Staffing follows the same model. The timely fashion. In addition, supplying service bureau can deploy a robust set of redundant servers and high throughput operations, technical engineering, and systems for the entire customer base will support personnel spread across the entire help to ensure high availability and base of small cable systems. Because the economies of scale. staffing for a centralized approach equates to a small fraction of an FTE on a per system Services from the centralized service basis, a top-notch staff can be assembled to bureau will require downstream connectivity provide superior services. from the service bureau via a satellite or terrestrial fiber-based data feed. The In order to address the unique needs of upstream requirements from the head-end to each cable system, it will be important for the service bureau can be managed using a the service bureau to identify and manage secure VPN connection. Cable operators each configuration individually. Additional that are currently offering High Speed controls will need to be in place in order to Internet services and/or VOIP services ensure the security and integrity of each already have the connectivity to the Internet cable system's data. that is required for this upstream data channel. Now, as a result of these For the service bureau approach to best investments in IP-based services, cable serve the customer base, it needs to balance system operators serving smaller markets the benefits of ubiquity with individuality. possess many of the prerequisites for using The goal of both separable or downloadable centrally-supported conditional access and security and OCAP is to offer a variety of OCAP services for expanding their lineup of choices into the marketplace. Offering a advanced digital video services. one-size-fits-all approach runs counter to the purpose of creating an open marketplace. Conversely, offering too many choices can increase the complexity of the operation and negatively impact the scale economics.

Conclusion

Cable system operators, CE viability depends upon being able to offer manufacturers and retailers, their customers customers the choice, convenience, quality and the regulatory community share a and ease of use that this next generation common vision for a fully open-cable content distribution platform can provide. A content distribution environment that centralized services bureau, which allows supports a wide array of devices and operators serving these smaller markets to applications. Achieving this vision for an outsource much of the capital and operating interchangeable world is a much steeper requirements of an OCAP infrastructure, challenge for cable operators serving rural represents an essential element in making markets. However, their future financial the interchangeable world available to everyone.

iCableLabs’ OpenCable initiative. See www.opencable.com iiCED magazine, January 2007 iiiCED magazine, January 2007 ivCableLabs’ OpenCable initiative. See www.opencable.com v“Don’t Leave Rural America Behind”, ACA, May 26, 2005 viFirst FCC Report and Order: Commercial Availability of Navigation Devices (PDF). FCC (1998-06-24). Retrieved on December 26, 2006 viiSonntagszeitung newspaper, March 18, 2007 viiiU.S. Federal Communications Commission document: "CS Docket No. 97-80: Report of the National Cable & Telecommunications Association on Downloadable Security" ixInteractive Momentum: Mike Robuck, Communications Technology, April 1, 2006 x NCTA - Key Industry Statistics 2006 xi ACA report to the FCC xii CED magazine, January 2007

Contact Information:

Gary E. Traver Senior Vice President and Chief Operating Officer

James A. Capps Senior Director, Software Engineering

Comcast Media Center 4100 E. Dry Creek Road Centennial, Colorado 80122 (303) 486-3800 www.comcastmediacenter.com

WiMAX As A Competitor To Fixed Line Broadband Technology Tim Burke Vice President, Strategic Technology, Liberty Global, Inc. Tony Werner Executive Vice President and Chief Technical Officer, Comcast Cable

Abstract WiMAX History and Design Goals

This paper examines the economic and To better understand the capabilities of technical capabilities of 802.16e WiMAX WiMAX let’s first describe the evolution of technology and contrasts it to fixed line the specification and its design goals. technology. WiMax was created out of an IP industry Technical capacity assessment of burst effort to replicate the success of WiFi rates, statistical load capabilities and technology but to do so in a controlled, wider performance under streaming and heavy Peer- area macro-environment instead of the to-Peer (P2P) environments are compared. It distance limited and unlicensed/unregulated also identifies frequency re-use plans and Radio Frequency (RF) environment of WiFi. antenna sectorization for varying subscriber The economic power of WiFi is its ability to densities. Performance analysis includes incorporate a standardized wireless broadband uplink and downlink performance and capability embedded into PC, laptop and throughput based upon the available portable computing devices. The incentives modulation, coding profiles and network and economics of this arrangement are configuration settings. compelling for CPE manufacturers, internet access providers and network operators, as the Various CPE is compared identifying the cost of the broadband wireless capability is trade-off of embedded CPE and network borne by the user. If WiMAX could utilize design. The cost and performance trade-offs the same economic vision of embedded for mobility are also defined. wireless broadband CPE while offering a secure, simple to access method similar to Economic analysis takes industry averages today’s wide area cellular networks it would and builds an end-to-end per subscriber cost surely come out a winner. In addition, the analysis based upon various homes passed belief that wireless technologies were overly densities, in building coverage, subscriber voice centric, narrow band, circuit switched speeds and mobility assumptions. The costs and solely focused on high speed mobility include site acquisition, construction, base created a window of opportunity for a station, backhaul, CPE and installation costs. technology with roots in the IP and fixed broadband data world. The conclusion section contrasts these economic and technical characteristics to There were some wrong turns during the fixed line and identifies the areas of development of the WiMAX specification. opportunity for WiMAX to be competitive. Early claims aggressively cited high user capacity, over great distances (70 miles) when travelling at very fast speeds (70 mph) were possible by many simultaneous users.

Fortunately, the developers of WiMAX specification. To summarize, the key remained steadfast to their vision and in a very requirements are1: persistent manner continued to upgrade the ƒ High average sector throughput to specification and standard through numerous support > 1 GB /user/month revisions (802.16-2001, 802.16-2004, 802.16- ƒ High cell edge performance on 2005). Equally pragmatic was the philosophy downlink (> 1,000 kbps) and uplink (> of the developers to build upon core 256 kbps) technologies and specifications with a proven ƒ High performance uplink & downlink track record, such as IP, DOCSIS and from NLOS indoor locations wideband RF channel technology while at the ƒ Support high number of simultaneous same time incorporating newly developed users (>150 per sector) technical advancements. Foremost among ƒ Low latency to support user experience these enabling wireless technologies built into and real time applications (e.g. VOIP) the specification were OFDMA (Orthogonal ƒ QOS (Quality of Service) to support Frequency Division Multiple Access) differentiated services modulation, MIMO (Multiple Input/Multiple ƒ Full portability and nomadicity Output) antenna technology, advanced coding ƒ High speed mobility (< 120 Km/Hr). techniques such as HARQ (Hybrid Automatic

Request) and TDD (Time Division Duplex) duplexing methods. In retrospect, their design Description of WiMAX Specification decisions were foretelling as experts in the cellular world are using the same technologies The WiMAX specification is a double- now in their submissions for the 4G wireless edged sword. It has many flexible settings and standards. assumptions that allow it to be optimized for a variety of business needs. For instance, if Right from the start there were some coverage, mobility and symmetrical services inherent constraints with the WiMAX are desired then it can be configured to match specification that still exist today. The biggest those needs. Conversely, the network can be obstacle being the spectrum ranges in which optimized for high capacity and asymmetrical WiMAX currently operates. The original capacity. As engineering and product standard, 802.16-2001, was for the LMDS marketing personnel can imagine, the spectrum (24, 28, 38 GHz), which is only good numerous design “knobs” will be a huge for line of site microwave type applications. benefit for delivering appropriate services to The current 802.16-2004 (Fixed WiMax or customers. Unfortunately, the flexibility 802.16d) and 802.16-2005 (Mobile WiMAX offered by this capability comes at a cost, or 802.16e) standards are more appropriate for namely complexity and ultimately Non-Line-Of-Site (NLOS) applications as they interoperability issues. operate in 2.3, 2.5 & 3.5 GHz ranges. Unfortunately, these frequencies are The biggest economic trade-off for a disadvantaged when trying to match the network operator is the decision to optimize propagation characteristics of cellular for either capacity or coverage. The flexibility frequencies (.8, .9, 1.8, 1.9, 2.1 GHz). of WiMAX gives a network operator the opportunity to optimize the economics of a Despite these challenges the developers of network design for either of these key criteria. the WiMax standard set out ambitious design Capital constrained start-up operators, without goals for a combination fixed broadband and the benefit of a customer base, may opt to mobile service prior to building the initially design a coverage based network that minimally meets the capacity requirements.

As the customers come online additional means the same frequency is used in capacity can be added with the appropriate every adjacent sector/cell) incremental capital. The power of the ƒ Sectorization (Omni, 3 sector, 4 WiMAX specification is that this incremental sector… 6 sector) capital can be minimized because of the high ƒ Downlink to Uplink Ratios of 1:1, 2:1, capacity capability of the standard relative to 3:1 other wireless standards. ƒ Sub-channelization techniques ƒ Channel bandwidths of 5, 10, and 20 A network operator must first make the MHz traditional business decisions of homes passed (addressable market), anticipated penetration Antenna Technology7,14 rates, desired product speed, and applications. o MIMO (where there are multiple At that point a critical decision for the operator transmit and receive antennas at of a wireless network, like WiMAX, is the both the CPE & Base Station) choice of terminal devices offered to the o Beam forming Antennas (lock in customers and the cell edge uplink data and track the CPE and null out speeds. Outdoor fixed antennas, indoor interference). gateways for PC’s, PCMCIA cards for laptops, laptops with embedded CPE and Unfortunately, optimizing these parameters handheld/mobile devices offer a variety of can be a “zero sum game” for the operator. market opportunities and different Improving one parameter can have a negative applications. For the network designer each effect on others. Particularly disconcerting is device has a unique link budget that weighs the affect of optimizing for coverage and/or heavily on the service speeds and network mobility which can severely affect capacity costs. The limiting item for a wireless design and vice versa. As is common in such trade- is typically the uplink data rate at the cell edge. off’s they typically result in economic choices. The cell edge is usually indoors through many For instance, spending more money per cell walls and far away from cell site. The linkage site for sophisticated electronics associated of the choice of CPE and uplink speeds at the with advanced antenna technology that gives cell edge is a critical starting point for the operator coverage gains versus just adding determining the network design and how the additional cost and simple base stations/sites. network should be best configured. Since WiMAX is a 4th generation (4G) In the WiMAX specification there are a wireless technology it is useful to understand number of key parameters and assumptions how WiMAX OFDMA (Orthogonal that can be adjusted to accommodate either the Frequency Division Multiple Access) coverage or capacity needs of a carrier’s technology differs from the previous business plan8: generations of wireless technology. Figure 1 ƒ CPE selection and cell edge uplink illustrates the conceptual differences in the budget frequency, time and power dimensions of ƒ Adaptive modulation assumptions multiple access technologies2. Multiple access ƒ Frequency reuse of N=1, N=2, N=3… techniques divide channels or voice (where N represents reuse, such as N=1 conversations by either time, frequency or unique identifiers like codes.

Frequency Division Multiple Access (FDMA) Time Division Multiple Access (TDMA) Code Division Multiple Access (CDMA) Power Time Time Power Power User 7 Time User 15 User 23

User 1 User Use User 9 User 1 8 User 7 c r 8 code code User 16 User 6 code ode User 2 Use User r 3 User 5 code User 3 4 code User 2 code code User 1 code

30 20 Frequency KHz 0 5000 KHz KHz Frequency Frequency 2G: Digital Cellular (GSM) 3G: Wideband Cellular (HSDPA, EV-DO) 1G: Analog Cellular (AMPS) Figure 1: Conceptual Differences of Multiple Access Technologies

In Frequency Division Multiple Access (FDMA) each user receives a unique In OFDMA users share tightly spaced frequency as shown above in the 1st generation subcarriers and time slots. The diagram above analog cellular example. Advancements in shows over 1,000 separate 10 KHz subcarriers multiple access techniques in the late 1980s in a 10 MHz channel. These subcarriers are led to giving each user a unique time slot orthogonal to each other meaning they are within the bandwidth allocation and digitizing unique and non-interfering. A stream of data the analog voice with vocoders. Time Division from an individual user could be assigned or Multiple Access (TDMA) 2G digital cellular scheduled a variety of different narrow technologies such as GSM provided for 7 user frequencies and time slots time slots within a 200 KHz carrier. Finally, in the mid to late 90s Code Division Multiple The next sections will give an overview of Access technology reached maturation and the key aspects of the WiMAX specification was promoted as 3G wideband cellular by focusing in on the critical design wireless. CDMA users share time and parameters and design trade-off’s. frequency slots but employ codes that allow users to be separated by the receiver29. Contrasting these multiple access techniques CPE Selection and Uplink Cell Edge to OFDMA is shown in Figure 2. Performance

Orthogonal Frequency Division Multiple Access (OFDMA) Each CPE device has different output Power Time power and antenna gains. Since the uplink is typically the limiting item for the link budget 1 er Us there is a strong relationship between type of

r 1 se device and cell site range. As an illustration, a U variety of different cell site ranges based upon 1 er Us the link budget for each of the various CPE 1 er 9 Us devices as shown in Table 1.

10 KHz

10 Frequency ,000 KHz 4G: Broadband Wireless (WiMAX, LTE) Figure 2: OFDMA

Antenna Transmit Cell By having adaptive modulation, the base CPE Type Gain Power Range stations and subscriber stations are able to Indoor 7 dB 27 dB 2.1 Km select at any given time the modulation rate separately for the uplink and downlink which PCMCIA -2 dB 24 dB 1.0 Km will yield the optimum operation given the current link conditions. Chart 1 is an Laptop 2 dB 27 dB 1.40 Km illustration of the coverage versus capacity Handset -2 dB 20 dB .8 Km tradeoff of using adaptive modulation schemes17, 18. At high modulation rates Table 1: CPE Devices & Cell Site Ranges (64QAM) the sector capacity is very high (10 Mb/s) but the range is low (< 1 Km). For simplicity, not all the individual elements Likewise at the lower modulation levels a and assumptions to the link budget are listed great distance can be achieved but only at the above. For example, in-building margins and expense of sector capacity. fade margins could vary by application and CPE. Additionally, propagation models vary Downlink Throughput & Distance Varies With Modulation widely for high speed mobility and fixed applications and can have a profound effect on 12.00 3.50 cell radius calculations as shown in Table 1. Throughput (Mb/s) Distance (km) 10.00 3.00 As one can see in Figure 3, coverage will vary 2.50 by CPE. This is mainly a result of the 8.00 probability of using a higher speed modulation 2.00 technique (e.g.- 64QAM) at the cell edge with 6.00 higher gain CPE. Since modulation and 1.50 4.00 coding modes are the key for performance, the 1.00 (km) Distance next section will describe how modulation is 2.00 uniquely used in the WiMAX specification. (Mb/s) Throughput 0.50

0.00 0.00 Residential Handset PCMCIA Embedded Outdoor Laptop Gateway Antenna 64 QAM 3/4 64 QAM 2/3 16-QAM 3/4 16-QAM 1/2 QPSK 3/4 QPSK 1/2 Modulation

Chart 1: Speed & Distance by Modulation

Switching between modulation rates and .8 Km 1.0 Km 1.4 Km 2.1 Km ~ 4 Km coding is based on the Signal to Noise Ratio Figure 3: Coverage by CPE (SNR). Typically, high modulations are achievable for locations near the base station while cell-edge regions with low SNRs due to Adaptive Modulation distance and potential interference from adjacent cells use low modulations. Fading Adaptive Modulation is one of the critical due to distance causes subscribers at cell-edge characteristics of the WiMAX specification. to operate at low modulation rates which do Adaptive modulation techniques allow an not require high SNRs. In the presence of individual subscriber to operate at different adjacent cell interference, the same subscribers modulation rates than an adjacent customer. may not be able to meet SNR requirement, and

hence, results in lower modulation modes. ƒ High modulation rates are selected Generally, NLOS environments suffer from when link conditions yield high SNR low SNRs due to reflection losses, diffraction resulting in higher performance. losses, multi-path fading and fading due to obstacles’ scattering effect. The ability to Figure 4 illustrates the concept of how the achieve high modulation at cell edge depends uplink modulation rate (QPSK ½) at the cell highly on the ability to maintain high SNRs edge is typically the limiting design item and given the demanding link conditions of a sets the speed of the user experience (256 NLOS environment. Table 2 summarizes Kb/s). Adaptive modulation, power and typical downlink WiMAX modulations and antenna gain of the CPE devices are key SNR levels supported4,9,10. drivers of product speeds, cell capacity, maximum subscriber counts and 3.4.16,17 Downlink Modulation SNR Required coverage . and Coding (dB)

QPSK ½ 1.8 QPSK ¾ 4.0 Cell Edge 16 QAM ½ 6.8

16 QAM ¾ 10.4 Up Link

64 QAM ½ 12.9

64 QAM 2/3 14.2

64 QAM ¾ 16.5 64 QAM 5/6 20.6 Table 2: SNR By Modulation

Without adaptive modulation, there are Figure 4: Uplink Cell Edge Representation two likely situations which can be experienced At this point it is important to briefly by subscriber stations: touch on the topic of WiMAX capacity in the ƒ Subscriber stations have high adaptive modulation discussion since detailed modulation rates even when link capacity calculations will be covered in later conditions yield low SNRs, resulting in sections. WiMAX average throughput high BER, and degraded performance capacity calculations are a controversial topic ƒ Subscriber stations have low as there is little agreement on those numbers modulation rates even when link because there are so many varying conditions yield high SNRs, hence, assumptions possible. Suffice it to say, a key resulting to inefficiency. aspect of the uplink and downlink WiMAX capacity claims is the assumption on which Under both scenarios, the end result is low modulation and coding scheme is used by the performance. By adaptively switching from customers. For instance, peak or theoretical one modulation rate to another, the system is 9,18,19 WiMAX capacities assume there is a single able to ensure that : user in a sector and they are operating at the ƒ Low modulation rates are selected best modulation and coding rates in both the when link conditions yield low SNR, 5 downlink (64QAM /6) and uplink (16QAM resulting in low BER ¾) 100% of the time. As can be imagined, high capacity claims can result across some vendors as a result of this assumption. The

realism of all users always attaining the highest modulation rate is questionable, which 1 is why we add the words theoretical and single 1 user when using peak capacity numbers. More 1 3 2 appropriate capacity claims are under the 3 2 heading of average throughput or average 3 2 1 channel capacity. Here again the adaptive 1 modulation distribution assumption is 1 3 2 absolutely critical. For example, the 2 distribution assumption of how many 2 1 3 subscribers (or the % of subscribers) operating 3 in each adaptive modulation mode drives the 3 2 overall average channel throughput number. Frequency Reuse Interference The concept of frequency reuse is integral to any wireless and mobility technology. Reuse margin Without this capability the wireless industry would never have matured in a scarce DL UL spectrum environment. Basically, in a three sector cell site shown in Figure 5 each sector N=1 5 dB 4 dB uses a different frequency. Cellular technologists call this an N=3 reuse pattern. N=3 1 dB .5 dB

1 Figure 6 & Table 3: Frequency Reuse Margins

The general conclusion that can be drawn from a higher reuse pattern is there is less interference from adjacent cells occurring with 3 2 an N=3 reuse over an N=1 pattern. Lower interference means a higher SNR (Signal to Noise Ratio) which translates into better Figure 5: Frequency Reuse (N=3) modulation formats and higher throughput or coverage. Likewise, the link budget is better This means three different frequencies are (lower interference margin) in a higher reuse used per site. More spectrum is used for N=3 pattern. Table 3 shows interference margins 3,4,7,8 frequency reuse (3X) than an N=1 for various reuse patterns . A 4dB configuration. Therefore the spectral difference between an N=3 and N=1 reuse for efficiency (# of bits per bandwidth in Hz) is WiMAX OFDMA environments is typical. worse in an N=3 than a N=1 reuse pattern. CDMA technologies can use a N=1 Figure 6 illustrates the N=3 frequency reuse frequency reuse because channels are concept with more than one cell site29,30. separated by codes not frequencies. OFDMA systems, such as WiMAX, have that luxury only at a capacity and coverage cost. OFDMA can approach N=1 frequency reuse but still

must combat adjacent cell or sector High data rates are attained in OFDMA interference. systems because the information rate is Fortunately, sub-channelization transmitted in parallel over a large number of techniques in WiMAX OFDMA systems can sub-carriers (1,024 in a 10 MHz channel)1,8,11. combat N=1 frequency reuse interference. In For instance, the CPE’s data stream is divided reality, all WiMAX vendors recommend the into several parallel streams of reduced data use of frequency reuse patterns greater N=1 as rates and each sub-stream is modulated and this provides the optimum trade-off of spectral transmitted on separate orthogonal (unique) efficiency, capacity & coverage. Again, a subcarriers. The high level diagram in Figure word of caution is needed here when looking 8 illustrates this concept for the transmit at WiMAX spectral efficiency (bits/Hz) claims portion of a CPE device. as N=1 frequency reuse assumptions are always used. Although N=1 is technically Sub-Carrier feasible the only way to get that reuse pattern and still attain claimed capacities (average Sub-Carrier Output Data sector throughputs) is to reduce the cell site RF Multipath Input Data WiMAX Modulation Stream over Channel radius to a very small distance. The frequency CPE Stream OFDMA 1024 OFDMA & Coding 10 MHz reuse, spectral efficiency, coverage tradeoff Sub-Carriers Channel just described is a perfect illustration of the “zero sum” nature of wireless technology Sub-Carrier when it comes to capacity versus cell range versus spectrum utilization. Figure 8: OFDMA Subcarriers

Sub-Channelization Techniques A key enabler in the WIMAX OFDMA system is the MAP (Media Access Protocol) One of the most enabling core technologies scheduler that allocates bits across various in the WiMAX specification is the concept of time slots and sub carriers (frequencies). In sub-channelization. Dividing the overall effect, this function is a very smart control channel (e.g.- a 10 MHz channel bandwidth) channel that obtains feedback on the channel into sub-channels used only by certain quality for each user and then tightly schedules subscribers on the uplink improves overall cell and packs user traffic in the optimum time and range and uplink capacities tremendously. frequency2,3. OFDMA sub-carriers (shown as arrows in Figure 7) are grouped to form Sub-Channels. Channel Bandwidth and TDD (Time Division Duplex) Frequency 10 MHz Channel Bandwidth MHz Channel 10 Data TDD (Time Division Duplex) sub-carriers technologies utilize the same slice of spectrum (tones) for both uplink and downlink communication. A core advantage of TDD is the ability to

Pilot allocate more of the available capacity to the Sub- downlink than the uplink which is particularly Carriers useful for asymmetrical data traffic. An overall high channel capacity can be obtained if a very high DL/UL ratio is assumed. In fact, Time most WiMAX peak or theoretical capacity Figure 7: WiMax OFDMA Sub- claims assume the highest 3:1 ratio as this Channelization

optimizes for capacity. Unfortunately, there is Average Sector Throughput (Mb/s) a coverage penalty for choosing such a TDD Ratio Downlink Uplink asymmetrical ratio. Figure 9 illustrates the (DL/UL) Capacity Capacity trade-off possibilities of TDD spectrum and 1/1 5.57 4.76 WiMAX technology. An operator can choose to maximize for coverage and provide less cell 2/1 7.28 3.30 sites by adjusting the DL/UL ratio16,17. 3/1 8.25 1.99 5 MHz Channel Bandwidth, with 2x2 MIMO Trade DLCapacity Table 4: DL/UL Ratio’s

For The technical details on why this effect ULCoverage occurs goes back to the concept of sub- Higher channelization and the subcarriers in OFDMA Power technology6,8,11. The UL coverage gain in going from a 2:1 ratio to a 1:1 TDD Ratio (DL/UL) occurs because with a 1:1 ratio, there are more uplink symbols (time slots) available .Figure 9: Dl/UL Trade-off’s for a subscriber data rate (256 Kb/s). Specifically, 21 UL symbols versus 15 UL Configuring a 1:1 DL/UL ratio gives 1.5 symbols. The fewer tones (data sub-carriers) dB of link budget improvement for the needed to be allocated in the uplink results in a 10 uplink over a 2:1 ratio. The higher power reduced uplink signal bandwidth (670 KHz from the Base Station makes up the loss in DL versus 930 KHz). The same power across a link budget. Figure 10 illustrates this smaller bandwidth results in an improved phenomenon when a coverage versus capacity sensitivity. This means the uplink coverage 5 calculation is performed . For a new entrant improves while maintaining the subscribers operator without customers it makes sense to uplink data rate of 256 Kb/s. The frequency initially deploy the network with a coverage and time domain representation in Figure 11 optimized TDD DL/UL ratio of 1:1 and incur illustrates this concept9,12,13. the capacity “hit” and then change to a more capacity optimized DL/UL ratio later on. TDD Ratio 2:1 TDD Ratio 1:1 Table 4 shows the DL and UL capacities for Frequency each of the possible TDD ratios 9,10. DL UL DL UL

DL/UL of 2/1 DL/UL of 1/1 Assuming In-Door CPE, 256 Kb/s UL speeds at cell edge 10 MHz Channel Bandwidth UL Signal DL UL Signal DL U Bandwidth UL Bandwidth L +10% 930 KHz 670 KHz

28 DL 15 UL 22 DL 21 UL Time 1.50 Km 1.75 Km symbols symbols symbols symbols Figure 10: DL/UL Representation Figure 11: Details of DL/UL Trade-off’s

Channel Bandwidth Selection coverage when going from a 5 MHz to 10 MHz channel bandwidth11,12,30,31. WiMAX is unique from previous wireless standards in that it allows for flexible or DL Link Budget loses 3 dB when increasing from 10 to 5 MHz Channels scalable channel bandwidth choices by the operator while maintaining the same interoperable specification. The ability to 5 MHz have the same CPE operate in either a 5 MHz Down Link or 10 MHz channel is extremely valuable as the network can grow in capacity without 10 MHz adding equipment or changing out CPE. Although the RF front-end hardware of current Up Link WiMAX CPE and Base Stations are not Figure 12: Channel Size Trade-off’s capable of handling 20 MHz channels the specification is capable of growing to a 20 Average Sector Throughput (Mb/s) Throughput MHz channel size. Interestingly the 4G Channel Downlink Uplink improves cellular specifications being created (called Bandwidth Capacity Capacity when LTE for Long Term Evolution) envision a 20 5 MHz 7.28 3.30 increasing MHz OFDMA channel 1,3. channel size 10 MHz 13.81 6.71

from 5 to 10 2/1 DL/UL TDD ratio with 2x2 MIMO A non-intuitive concept of OFDMA and WiMAX that is foreign to the world of CDMA Table 5: Channel Size Capacity is the concept of being able to increase capacity by increasing the channel bandwidth Advanced Antenna Technology without losing coverage. Increasing from a 5 to 10 MHz channel adds considerable network Because WiMAX utilizes TDD and capacity to a sector (cell). Either greater OFDMA technologies it has some inherent speeds can be provided to the same number of advantages when it comes to smart antenna subscribers or more subscribers can be served technology. In TDD operation the same RF at the same product speeds. Increasing from 5 channel is used for both transmit and receive to 10 MHz bandwidth causes a 3 dB downlink so the RF channel link conditions at any point loss in the link budget3 as shown in Figure 12. in time are known. This allows for fast, If the full channel is used then less BTS power closed-loop type adjustments by the base is applied across the larger 10 MHz channel. station and CPE that can increase There is no loss in the link budget for the up performance. Additionally, OFDMA is not as link because of sub-channelization. The susceptible to frequency selective fading as other technologies, which plays to the bandwidth of each sub-channel is the same 6 regardless of the total channel bandwidth strengths of advanced antenna techniques . In therefore the same amount of power from the the specification WiMAX has many smart CPE is applied across the uplink bandwidth. antenna technologies incorporated into the Since the uplink is usually the most limiting standard as options. There are so many item for coverage there is no reduction in cell options and unspecified vendor range when WiMax capacity is increased (as implementations that it adds a level of shown in Table 5) by going from 5 to 10 MHz. complexity to the standard that will surely In the rare situation that the design was cause many interoperability issues with CPE downlink limited the network would lose devices. If the interworking aspects can be resolved over time some extremely beneficial capacity and coverage gains will be reaped

with these technologies. There are various keep the subscriber unit costs low. Figure 13 technologies that focus on either improving illustrates the typical WiMAX 2x2 MIMO cell site range such as Adaptive Antenna configuration downlink and 1x2 MIMO Systems (AAS or Beamforming) or increasing uplink. capacity such as MIMO (Multiple Input Multiple Output) Spatial Multiplexing (SM) DL: 2x2 MIMO technology. UL: 1x2 MIMO

Increasing capacity is the main benefit of MIMO. MIMO can improve capacity by 1 Tx / 2 Rx transmitting parallel data streams using multiple antennas at both transmitter & receiver. For example, if there are two 2 Tx / 2 Rx transmit antennas then each will carry ½ of the total data in the same spectrum at the same time. Likewise the receive antennas demodulate and combine in the same way. Therefore twice the capacity is possible.

MIMO uses multipath to its advantage and Figure 13: 2x2 MIMO works best in urban environment where the signals transmitted by the antennas bounce off Adaptive Antenna Systems (AAS) or buildings and take many paths before they Beamforming improve link budget & reduce reach the multiple receive antennas. If the interference. Several antennas (at the base received signals are uncorrelated they can be station only) form large directional arrays with combined in many ways (e.g.- Maximum a narrow beam width. The higher antenna Ratio Combining or MRC) to increase gain improves range and the narrow beam performance. If two transmit antennas are at reduces interference to and from other cells. the BTS and two receive antennas at the CPE 7,14 In the same way signals are added to improve it is called 2x2 MIMO in the downlink . gains, unwanted signals or interference can be subtracted to create a null directed towards MIMO capable CPE will have two receive interferers and thereby reduce interference. antennas and one transmit antenna while base Finally, Adaptive MIMO Switching or AMS is stations will initially have two transmit and the technology that adjusts the downlink to two receive antennas evolving to 4 transmit choose the best advanced antenna option for and receive antennas on the base station side. the current RF conditions. For instance AMS If the CPE and Base Station have only one can dynamically switch from a MIMO transmit and one receive antenna then it is capability to an AAS mode dynamically6. called a SIMO or SISO (Single Input Multiple Because of the complexity of AAS, MIMO Output or Single Input Single Output) and AMS these capabilities will be proven out configuration. It will take many years to get in during later phases of the WiMAX industry CPE prices low enough to accommodate two certification and interoperability testing. transmit antennas on the uplink from the CPE. WiMAX Capacity Claims WiMAX will be using mainly 1x2 and 1x4 MIMO capabilities on the uplink. MIMO The WiMAX Forum and other vocal transmit diversity puts the extra antenna at the proponents of WiMAX technology typically base station instead of the CPE in order to state peak data rate capacities. Although

interesting numbers these claims are very A large contributor to higher capacity theoretical as they represent the peak rate a numbers is the type of traffic model used in single user operating in perfect conditions can the simulation. For example, a simulation that obtain at the highest modulation and coding sends 100% large packets (called full buffer) rate. In addition, MAC (Media Access will have a much smaller MAC layer overhead Control) layer overheads for functions such as count therefore making the average throughput resource allocation and access controls are not capacities larger. If more realistic call & taken out of the number. Overall, stating peak traffic models are used that represent a mix of rates and comparing them with other wireless large and small packet type traffic (such as and fixed technologies is not very useful. web browsing) the overhead counts will be Table 61,5,23,24,26 lists the aggressive WiMAX higher and result in lower net throughput Forum capacities on the far right column for a capacities27. 10 MHz channel, 3:1 DL/UL ratio and N=1 frequency reuse configuration. Peak, average Additionally, a lower gain, less complex sector throughput and spectral efficiency type CPE receiver is used instead of the more numbers are listed for downlink and uplink in advanced non linear receiver assumed by the SIMO and MIMO environments. WiMAX Forum1,2,3,8. Certainly these advancements may be possible in future years The middle column illustrates much more but it is yet to be proven in commercial realistic assumptions that reflect the real world deployments. mobile wireless environment of today. As an example the 14.1 Mb/s MIMO downlink claim Lower power base stations that provide 2 by the WiMAX forum versus the 9.0 Mb/s to 4 watts at the antenna were assumed industry average is a 35% reduction in because the cost and size of a high power capacity due to changing assumptions. amplifier may be difficult to implement. The WiMAX Forum is assuming a 20 watts at the 1,8 WIMAX Capacity WiMAX Industry WiMAX Forum ground assumption which is a less practical Aggressive Assumptions N=1 reuse, 10 MHz BW, 3:1 DL/UL ratio More Conservative Assumptions RF implementation for tower configurations. Downlink Peak Data Rate SIMO 23 23 (Mbps)1 MIMO 46 46 The effect of higher power base station causes Uplink Peak Data Rate SIMO 3.5 4.03 a larger distribution of subscribers obtaining (Mbps) MIMO 3.5 4.03 the higher modulation rates which increases Downlink Avg. Sector SIMO 7.5 8.8 Throughput (Mbps) MIMO 9.0 14.1 channel throughput capacity. It is very Uplink Avg. Sector SIMO 1.1 1.38 important to realize that average throughput Throughput (Mbps) MIMO 1.5 2.19 Downlink Spectral SIMO 1.0 1.21 capacity claims assume a mix of modulation Efficiency (b/Hz) MIMO 1.2 1.93 modes. Uplink Spectral SIMO .40 .55 Efficiency (b/Hz) MIMO .60 .88

Table 6: Capacity Comparison

Affect of Changing Assumptions on WiMAX Downlink Capacity 50 46.0 Mb/s 45 40 35 30

25 23.0 Mb/s 20 -20% -15% -5% -10% -20% -25% +200% 15 14.1 Mb/s 11.3 Mb/s 9.6 Mb/s 9.1 Mb/s 10.0 Mb/s 10 8.2 Mb/s 6.6 Mb/s Downlink Capacity (Mb/s) Capacity Downlink 5 5.0 Mb/s 0 Baseline SIMO Baseline Realistic Call Low Gain CPE Low er Pow er SIMO Mobility 2:1 DL/UL ratio N=3 WiMAX Forum Antennas WiMAX Forum Traffic Model Reciever Base Station Antennas Propagation Frequency (No 2x2 MIMO) (No 2x2 MIMO) Model Reuse Peak Data Rates Average Sector Throughput (Net Information Rate) Baseline Assumptions WiMAX Forum Baseline Assumptions • Spectral • 10 MHz Channel • 10 MHz Channel efficiency is • 3:1 DL/UL ratio • 3:1 DL/UL ratio reduced by • N=1 Frequency Reuse • N=1 Frequency Reuse ~30% (from 1.9 • Single user in sector • Distribution of many users operating at various modulation & coding modes to 1.3 bps/Hz) • User operating at highest • MAC OH taken out of average sector throughput capacity number modulation/coding (64QAM 5/6) • Using fixed channel propagation model (0-3 Km/hr) • MAC OH not taken out • High performance receiver used at CPE • 2x2 MIMO • 2x2 MIMO • 20 watt power BTS at base of tower • Multiple users in sector using100% full buffer (large packet) traffic model

Chart 2: Impact of Assumption Changes on WiMAX Downlink Capacity Therefore, if one simulation assumed a when a 30 Km/hr channel propagation model majority of users get 64QAM versus a second 27,28,29 is used instead of a mix of channel simulation which assumed a lower percentage models skewed towards a fixed environment, of the users are able to obtain the highest (e.g.- 0 modulation rate then a lower capacity claim will result in the second simulation. to 3 Km/Hr used by the WiMAX Forum), Chart 23,5 gives a more detailed breakdown the anticipated gains and ultimately sector of how on a broad-guage percentage basis capacities are reduced16,17,22. varying the key assumptions mentioned above will affect the WiMAX downlink capacity Finally, if either a more symmetrical user numbers. Additional assumption changes that service or a network design optimized for may reduce average sector throughput from coverage is required then a 2:1 DL/UL (or the baseline WiMax Forum numbers are; no even 1:1) ratio will be required and impact MIMO antenna capability, using a full capacity. In fact, a subscriber average uplink mobility channel propagation model and service level of 256 kb/s combined with a reducing the downlink to uplink ratio. moderate VoIP mix of services and the need to WiMAX deployments with a simpler SIMO optimize for range will quickly drive the antenna scheme or perhaps rural applications operator to design to a 1:1 DL/UL ratio where little multi-path fading exists will because of uplink requirements25,26,27. In severely hamper the MIMO gains anticipated these situations a large overall average sector in computer simulations. Additionally, MIMO throughput reduction for the downlink will gains have not been fully proven when a user result as shown in the chart above. This is an is operating at low modulation rates such as illustration of a classic design trade-off where QPSK ½, which will be the case at cell edge uplink demands result in a downlink capacity operation6. Regarding the affect of mobility, “hit”.

On the positive side, extremely large capacity gains are possible by increasing the Good historical references from other new frequency reuse to an N=3 scheme and reduce wireless technologies, such as CDMA, exist. interference as shown above. There are When CDMA was first deployed the claims on tradeoffs to this design decision particularly, capacity and coverage were based on the negative effect on spectral efficiency. simulations. As commercial deployments Although the average sector throughput were rolled out it became apparent that many capacity increase is greater than 100% with an claims were either based on faulty N=3 frequency reuse, it also means three times assumptions, never achievable, or would take the spectrum is used. The spectral efficiency many years of fine tuning and enhancements bits per hertz ratio is upwards of 30% lower to reach. Over time (10 to 12 years), CDMA for N=3 over N=1 because of the much larger technology met its original claims and well total spectrum used9,16,30. exceeded the prior wireless technologies in capacity, coverage and performance. It is the One could infer from Chart 2 that a opinion of the authors that in the same way wireless operator with a large amount of OFDMA will surpass its predecessor wireless contiguous spectrum and a willingness to use technology. an N=3 frequency reuse and build a very dense network (< 1.5 Km cell radius) could offer WiMAX Subscriber Speeds and Services some reasonable downlink capacities. For instance, the 9.1 Mb/s sector capacity (before Translating WiMAX channel capacity MIMO was removed) could feasibly double to numbers we have just gone through into a a 15 to 17 Mb/s average sector throughput. useful understanding of the number of The operator would have to be willing to subscribers that can be served in a WiMAX forego offering mobility (change from a cell site and network requires standard traffic mobile to a fixed propagation model) and take engineering dimensioning. Paramount to this the uplink capacity restriction associated with analysis is an understanding of the likely a 3:1 DL/UL ratio. Although a WiMAX 20 product mix, and market statistics as shown MHz channel bandwidth is quite a bit down illustratively in Table 7 & 8. the road in developmental time frames, downlink sector capacities under similar These design assumptions are needed to assumptions might some day be able to determine the cell site radius, number of sites approach 25 to 30 Mb/s downlink. and sectors and subscribers served per sector. To summarize, channel bandwidth, Desired product speeds, market density and frequency reuse and downlink to uplink ratios subscriber penetration levels are particularly important starting points as is the overbooking are very big drivers of capacity. As noted in ratio (or concurrency) assumption. Since the Chart 2, MIMO antennas, base station power, small packet size and large header CPE receive sensitivity, traffic models and requirements of VoIP services are a burden on channel propagation model assumptions can capacity, the VoIP penetration level is likewise also skew results in a variety of directions. necessary. In conclusion, one sees many differing capacity claims regarding WiMAX. It is feasible that all these claims are the result of very accurate and sophisticated simulations, but are not necessarily comparable, (nor realistic) as the key underlying assumptions will most likely vary.

6.00 Component Assumption Relationship Of WiMAX Cell Radius To Customer Speeds For Various CPE Total Area 443 km2 5.00 CPE Indoor Gateway Fixed - Outdoor Fixed - Indoor Market Penetration 10% Laptop - Indoor 4.00 Handeld - Outdoor Total Homes in Area 186,000

3.00 Product (DL/UL)Speeds TakeTake OverbookingOBF Avg.Avg. Kbps (DL/UL) Rate DL/UL Kbps 2.00 1024 / 256 10 40 / 15 25.0 / 6.4 Cell Radius(Km)

512 / 128 40 25 / 15 20.5 / 5.1 1.00

256 / 64 50 20 / 15 12.8 / 3.2 0.00 Wtd. Avg 100 17.1 / 4.3 64 / 32 128 / 32 256 / 64 512 / 128 1024 / 256 1500 / 256 2048 / 256 3072 / 256 Design Target Customer Speeds (DL/UL) Assumptions: • 5 MHz Channel Bandwidth, 2.3 GHz frequency, 2:1 DL/UL ratio, SISO Table 7&8: Product Mix and Market • 50% Data Penetration, 10% Voice (VoIP) Penetration, 729E VoIP coder • Market Density: 420 HP’s per Sq. Km. Statistics Chart 3: Cell Radius to Customer Speeds

As customer speed requirements increase In the same way a graphical representation the network goes from coverage limited to of total subscriber capacity (Chart 4) in a 3 capacity constrained and the cell site radius is sector cell under the same varying customer reduced to meet the demand. Chart 3 shows speeds and CPE device scenarios10,11,12,16,17. the effect of desired customer speeds and CPE 13,16 Once again, SNR’s and adaptive modulation devices on cell site radius . In particular, rates required at the cell edge for both speeds in the uplink offered to customers can downlink and uplink are the critical severely impact the cell radius calculations. calculators. Given the desired customer The key drivers of this graphical speeds at the cell edge, customer product representation are the typical SNR’s and speeds and overbooking assumptions a target associated adaptive modulation rates required number of subscribers per cell and sector can at the cell edge for both DL and UL link 1,5,17,18 be calculated. budgets . It is easy to see that the link budget is uplink limited for all but the outdoor The conclusion of charts 3 and 4 is that a CPE case as cell radius is flat until the uplink WiMAX three sector cell site has the capacity customer speeds are increased. The design to serve approximately 125 customers per target for WiMAX networks is optimum at 1 sector (382 subs / 3) at 1 Mb/s downlink and Mb/s DL & 256 Kb/s UL for indoor CPE. 256 Kb/s uplink when the cell radius is 1.5 Km and the CPE is a fixed indoor residential gateway. Even the subscribers at the cell edge and operating at the lowest modulation rate will be able to receive a 1 Mb/s downlink service. As wireless system design is very probabilistic and uplink limited the modeling used here focuses on the probability of serving customers at the cell edge with 256 Kb/s uplink service.

4500 Although the average service level offered Relationship Of Subscriber Capacity To Customer Speeds ForVarious Various CPE CPE Devices to an individual customer is sized to 1 Mb/s 4000 downlink speeds certainly an individual user 3500 Fixed - Outdoor Fixed - Indoor could burst to the entire sector average Laptop - Indoor 3000 Handeld - Outdoor throughput number. It is unrealistic that a

2500 single CPE will have access to a fully unoccupied channel. Most of the current 2000 version CPE hardware is rate limited to 3 to 5 1500 Mb/s downlink speeds and it is reasonable to

1000 assume most network operators will tightly

Maximum Subscribers Per Cell Site Cell Per Subscribers Maximum control via software the downlink and uplink 500 speeds of wireless users. 0 64/32 128/32 256/64 512/128 1024/256 1500/256 2048/256 3072/256 6144/256 Customer Speeds (DL/UL) It is appropriate to reiterate in this section Design Target Assumptions: • 5 MHz Channel Bandwidth, 2.3 GHz frequency, 2:1 DL/UL ratio, SISO (no MIMO) that average sector throughput numbers, such • 50% Data Penetration, 10% Voice (VoIP) Penetration, 729E VoIP coder • Market Density: 420 HP’s per Sq. Km. as the 7 Mb/s number, is truly an average. If Chart 4: Cell Capacity to Customer Speeds an individual user is in perfect channel conditions and operating in a 64QAM adaptive It is certainly possible to serve either more modulation mode then that user could in customers at lower speeds or fewer customers theory attain a 10 Mb/s speed even though the at higher downlink speeds. Using the same average throughput is 7 Mb/s. Conversely, assumptions a 10 MHz channel would serve cell edge customers may not be able to burst between 200 and 300 customers per sector above their stated 1 Mb/s threshold service with the similar service levels. level.

Another interesting effect that can be Coverage Considerations concluded from these charts is if a dense network is initially built then downlink speeds The old cellular adage “coverage is king” from 1 to 5 Mb/s can be provided to certainly applies to WiMAX as well. It is a customers. The uplink on the other hand is particularly appropriate mantra in the initial extremely sensitive to increasing customer launch stages when the network has not added service levels and uplink sensitive VoIP type any customers and the operator is uncertain services. Optimizing for uplink is possible but where and when they will appear. only with a coverage or downlink capacity negative impact. There are many variables that affect coverage. Certainly the frequency and power As the penetration levels increase capacity limits set by the spectrum being used are will need to be added by either growing to a 10 foundational variables. Network and WiMAX MHz channel bandwidth, adding more sectors specification settings we have mentioned are and using more frequency, upgrading to critical. Base station antenna height, TDD MIMO CPE (mainly in urban areas), going to DL/UL ratios, frequency reuse, CPE type and a higher DL/UL ratio (at the expense of adding smart antenna have the most impactful new sites to cover the shrinking cell radius) or engineering design decisions. Customer cell splitting (adding new cell sites). If the propositions agreed to in the business case can above scenario wasn’t already at an N=3 vary cell site range quite extensively as well. frequency reuse increasing the frequency reuse For instance, uplink speeds available to would also help with both range and capacity subscribers at the cell edge, mobility versus increases. fixed services, probability of attaining coverage at the cell edge and in-building

versus outdoor coverage are key factors. All A choice of using either 3.5 GHz or 2.3 GHz ten of the variables mentioned above are baked spectrum is an easy decision, as the lower into the link budget along with the chosen frequency has better propagation. vendors receive sensitivities and transmit Additionally, a higher 25 meter cell site power gains of their equipment to arrive at a height, 1:1 UL/DL ratio and N=3 frequency maximum allowable path loss for both the reuse choices will optimize for coverage over downlink and uplink separately. Typically the capacity and spectral efficiency. uplink will be the most limiting so that path loss will be used to calculate the cell radius for Chart 6 provides a similar representation of a typical terrain (suburban, urban). Chart 5 the impact on coverage for various customer provides an indication of the degree each of impacting variables9,10,11,16,22. the technical variables affect cell radius9,10,11,16,21,22. Effect of Customer Based Variables on WiMAX Coverage 5 Effect of Technical & Network Variables on WiMAX Coverage + 260% (handheld to indoor)

4 4

Outdoor 3 3 + 28% (20m to 30m) + 50% (512 to 256 Kb/s) 30 m 2 64 Kb/s + 32% 25 m + 58% 128 Kb/s + 29% 80% Fixed Indoor 2 + 44% + 24% (3:1 to 1:1) 256 Kb/s Fixed + 26% 1 Laptop/PCMCIA 1:1 2.3 GHz Smart Antenna Low Speed 90% 2:1 (Km ) Radius Site Cell C e S ll ite R a d iu s(K m ) 20 m N=3 512 Kb/s Mobility Handheld 0 1 Standard UL Customer Speeds Mobility Coverage Probability CPE Ty pe 3:1 N=1 3.5 GHz Antenna @ Cell Edge

0 Frequency Cell Site Height TDD DL/UL Ratio Frequency Reuse Base Station Antenna Chart 6: WiMAX Coverage Impact - Customer Chart 5: WiMAX Coverage Impact - Technical Uplink speeds offered to the customer at the cell edge can cause widely fluctuating cell Going from a 3.5 GHz to a 2.3 GHz radius. Offering a service tier to customers at spectrum operation can increase cell radius 256 Kb/s instead of 512 Kb/s can increase cell approximately 44%, while a 10 meter antenna size by 50%. Cellular networks are typically height increase can easily provide a 28% designed for 64 Kb/s indoor coverage, which improvement. As mentioned in previous creates cell sizes twice the size of a 512 Kb/s sections reducing the DL/UL ratio from 3:1 to designed WiMAX network. 1:1 can grow cell site range from 1.4 Km to 1.74 Km. Frequency reuse and the associated Choosing a mobility based path loss gains in link budget, by reducing interference, channel propagation model introduces fading will increase range 26% for a 5 MHz channel margins in the link budget that can easily add at 2.5 GHz. Finally, the cell radius gains from almost 30% cell reductions. Building in for various advanced antenna technology varies overlapping cell coverage for handover also tremendously. A typical industry average can adds additional cells. provide a 58% improvement in coverage. Another subtle tool used by RF engineers Most new entrants opted for range when to increase range is to set the threshold of making decisions on the variables in Chart 5. reliability at the cell edge to a level lower than

the industry standard 90%. 80% coverage will take its toll on capacity. Both industry probability means 20% of all users at the cell simulations and testing confirmed some major edge will not get the minimum level of uplink uplink capacity reductions associated with service required by their service tier (e.g.-256 small size packet call traffic over full buffer kb/s). type traffic1,2,3,27.

As mentioned in the very beginning of the Packet loss, bit error rate (BER) and packet paper the choice of CPE devices drive a wide error rates (PER) as the system is loaded and variance in cell radius. In a 2.5 GHz design, stressed will be important performance issues the cell radius can grow from 1.76 Km in a to watch as the technology develops and suburban area for indoor CPE to 4.65 Km if commercial networks are deployed. To the the CPE is mounted outdoors. Rural terrain extent the complicated MAP scheduler has an even larger effect, as a 2.32 Km indoor function operates and achieves the optimal configuration grows to 6.55 Km with an performance of the WiMAX specification will outdoor installation. determine the BER stability.

Wireless broadband cell coverage Latency could become an issue with engineering certainly contains a large amount WiMAX as the inherent nature of TDD of trial and error and has many elements of art systems causes a longer round trip delay as well as science. The end result is a associated with the transmit and receive paths somewhat variable end product offered to sharing the same swath of spectrum. Industry customers. Some subscribers will have an testing to see if actual performance matches over- engineered service while others in the anticipated 50 ms specification8,25,27 will be difficult terrain or well inside a building made a closely monitored performance data point. of difficult materials are unknowingly given a poor level of service. As mentioned in the previous section, the variability of product speeds (both downlink Performance Considerations and uplink) seen by customers in the real world will be a marketing and technical WiMax will have many performance challenge for operators. Predictive tools will issues as the technology becomes developed help to proactively solve these issues but there and deployed. The most challenging aspects will be no substitute for having ancillary CPE will be uplink related performance. such as window mounted antennas and higher power devices to assist the operator in a Maintaining uplink speeds and capacity at variance of customer service levels. the cell edge with small packet size applications such as VoIP will be challenging. Mobility will have its own share of OFDMA technology performs much better performance issues and will be left for the later than CDMA or TDMA schemes because of stages of WiMAX development. The sub-channelization. By splitting the entire 10 specification focused on the MAC and PHY MHz channel bandwidth across many layer so handover and other key mobility subscribers each customer uses only a small functions will require greater definition in the subset of subcarriers with a far lower power core network to reduce interoperability issues. than if it had to transmit over the entire MIMO technology has yet to be proven in 3,4,5 bandwidth . Even with that advantage the the field. In particular, much is riding on the uplink capacity is an extremely limited capacity gains associated with this feature. resource and inefficient small packet Very visible industry tests will be made to applications such as VoIP and web browsing

measure the gains associated with suburban design parameters. Wireless engineers are and rural terrains, where minimal multipath good at either optimizing a network for either exists, and to understand the gains with mobility or separately optimizing for operation in lower modulation and coding broadband capacity. It will be a new and modes. challenging task to optimize for both objectives. From a cost perspective it may Interoperability will be the key become an uneconomical design as there will performance issue. The WiMAX specification be a requirement to design for the lowest has many options, settings and specification common denominator. Mobility propagation vagueness subject to vendor interpretation and models, cell overlap (for handover) and high implementation. Overall the complexity and speed fixed uplink services inside the home choices comprised in the standard will will all trigger a very dense network design. invariably lead to substantial interoperability challenges. Particularly challenging is CPE to Secondly, the fact that a wireless network base station inter-working. Initial WiMAX has a very tight linkage between the uplink and chipset vendors are small companies and their downlink means you can not treat the coordination with many large base station engineering and design of these two segments vendors could unfortunately lead to design separately. Optimizing for the downlink will errors in the world of a specification with surely have an effect on the uplink and vice many choices. versa. Typically, broadband networks that can separate these two dimensions are easier and Finally, it will be important to manage peer more economical to manage performance to peer traffic, especially in the uplink, as is levels. common in the fixed broadband networks. Chart 7 depicts the capacity gains possible by Building on this second point, performance closely managing a WiMAX network uplink in wireless networks is all about tradeoffs. resource. The chart indicates moderate to The classic analogy for wireless is that aggressive peer to peer management can raise performance management is a fixed triangle, overall subscriber capacity by 30%. where coverage is at the apex, capacity is at one base and quality (or service levels) is at 500 P2P Management Impact on WiMAX Uplink Capacity the third corner. Optimizing for one of the 400 three corners always puts pressure and reduces the capability on one or both of the other 300 corners. Non-negotiable trade-offs result in 200 such an environment where there is always a losing metric with regards to performance. 100

Subscribers Per Sector Per Subscribers 0 No P2P Moderate P2P Extremely Aggressive WiMAX Economics Management Management P2P Management Peer to Peer Management The relationship of capacity, coverage and Assumptions: 10 MHz, 2:1 DL/UL, 3072 Kb/s DS, 256 Kb/s US, 3 Sectors, 20% data & voice penetration, 440 HP/Km2 density, 60% DL & 85% UL P2P Traffic performance to economics is also a very important trade-off. One of the main drivers of Chart 7: Peer to Peer Management WiMAX economics is the penetration rates and market density of the customers served. In conclusion, maintaining a high level of As Table 9 indicates, wireless networks with performance in a wireless broadband network small cell radii, such as WiMAX, the linkage will be challenging. Mobility and broadband have been historically two mutually exclusive

of market density and penetration levels to the subscribers served is critical2,10. WiMAX Capital Expenditure Components Per Site

200 WiMax Subscribers Per Site For Various Market 180 160 $50K $145K Densities & Subscriber Penetration Levels 140 $60K Market E. Europe Country US Metro Area Asia Suburbs W. Europe City 120 $30K $70K 2 100 Area in Km 20,273 12,949 443 830 $30K $75K Homes Passed (HP) 509,096 983,000 186,000 425,000 80 $30K 60 $10K $30K

2 C ap ital (U S D $000) $45K $20K Market Density (HP/Km ) 25 76 420 512 40 1.0% 1 3 15 18 20 $45K es t 3.0% 3 8 45 54 0 a Base Station Backhaul, Batteries & Total Base Station Construction Costs Tow er Costs Total Site Costs (Incl. Allocations) Environmentals (Collo, Rooftop or Tow er) (Greenfield Only) (Construction Blend) R 10.0% 9 27 223 272 802.16e BTS Microwave Site Acquisition 20% Greenfield 3 Sector, 10 MHz (DS-3 or 100 Mb/s) Legal 80% Rooftop or Collo on 15.0% 14 40 335 408 DL/UL; 2/1, N=3 reuse IP router/ switch Construction/Civils

ti Redundant Cards Installation Rigging 16.6% Outdoor Cabinet Cabinet w/ Electrical

ra 15 45 370 452 Transportation Environmentals Power

t Installation, Test & Turn-up Battery Back-up (4 hrs) Miscellaneous 20.0% 18 54 446 544 BTS Controller & NMS Allocation ene 25.0% 23 67 558 680 P Chart 8: WiMAX Capital Cost Management Table 9: Subs Per Site for Various Market Densities It is important to realize that the costs shown in Chart 8 can vary widely based upon Typically cities with market density 2 the vendor’s equipment and design. Certainly greater than 400 HP/Km such as the Western the type of antenna, the allocation for the base European and Asian suburbs shown in Table 9 station controller function (called ASN or have the best per site subscriber levels. Access Services Network in the standard), Because WiMAX has the potential to offer redundant cards, spares and software very reasonable channel and subscriber feature/functionality will drive a wide range of capacities (compared to other wireless costs. technologies) at short ranges, high density/high penetration conditions are very Similarly, the construction assumptions conducive to economic success. can cause a large cost variance. Foremost in the construction economics is the degree an This point is best proved by first looking at operator can forego the cost of building towers the WiMAX network cost structure. As Chart by obtaining collocation agreements with 8 illustrates the main costs associated with the existing tower owners and landlords of upfront network build are construction 2,19,21,23,32,33 rooftops. For the purposes of Chart 8, a 80% / related . Typically, site acquisition, 20% split of collocations to “Greenfield” new legal, civil works, tower costs, power & builds of towers is assumed. electrical costs are almost 50% of the total costs. This is true of most wireless networks One important learning we obtained in the but is amplified in a WiMAX network because understanding of the cost components and how the actual base station costs are reasonable. to minimize expenditures was the need to The second largest expenditure is for all the minimize construction costs by obtaining a costs associated with the base station low-power, small footprint base station and its electronics. When the microwave, installation, ancillary equipment. If large base station and rigging (cabling on a tower or rooftop) test & microwave backhaul electronics are required turn-up, battery back-up, environmental (A.C. then large shelters, air conditioning, additional or heat exchanger), cabinets and ancillary power and larger battery back-up gear is equipment are added together they account for needed. All these things drive higher 30% of the total costs.

construction costs as now cranes are needed, this number. Table 10 provides a variety of additional electrical work from the power cost per sub numbers for the illustrative company and larger concrete footings will be markets already discussed2,10. triggered. In the harsh environmental conditions of some rural networks the payoff WiMAX Costs Per Subscriber For Various Market of having “hardened” microwave and base station electronics is substantial in order to Densities & Subscriber Penetration Levels avoid these add-on costs. Market E Europe Country US Metro Asia Suburbs W Europe City Area (KM2) 20,273 12,949 443 830 Applying the up front Capital costs to the Homes Passed 509,096 983,000 186,000 425,000 market density and penetration numbers HP/KM2 25 76 420 512 discussed previously will lead to a cost per es t 1.0% $145,225 $48,558 $9,892 $8,281 subscriber metric shown in Chart 9 a 2,19,21,23,32,33 R 3.0% $48,558 $18,350 $3,447 $2,910 . Key to these calculations is the 10.0% $16,336 $5,595 $529 $475 on capacity, traffic engineering, coverage and ti 15.0% $10,582 $3,850 $428 $391 ra customer performance metrics illustrated in t 20.0% $8,281 $2,910 $377 $350 prior sections. The resultant cost per sub is

ene 25.0% $6,529 $2,389 $347 $325

$550 where over 40% of the costs is for CPE. P Constructions costs contribute 28% and the Assumptions: 1.90 Km cell radius, 3 sector, 10 MHz channel, N=3, 2:1 DL/UL ratio, 1 Mb/s DL & 256 Kb/s UL, Fixed Indoor CPE base station, microwave and ancillary Table 10: Cost per Sub for Various Market equipment split the remaining 30%. The cost Densities structure assumes a dense 1.3 Km cell radius, 2 moderate market density (440 HP/Km ) and Not surprisingly the cost per subscriber penetration level (20%). Finally, the network numbers begin to look quite beneficial for was designed for 90% coverage at the cell highly dense markets where operators believe edge with 1 Mb/s downlink and 256 Kb/s they can secure many customers. For uplink service levels. example, a Western Europe city with over 500 homes passed per square kilometer can reduce WiMAX Cost Per Subscriber the cost per subscriber metric by 20%. .

600 $225 $550 550 500 Another spin on this sensitivity is to look 450 at the same metric while just varying the cell 400 radius assumption10,12,13,16 as shown in Table 350 $152 $9 $325 300 11. For instance, 128 Kb/s UL speeds with 250 80% cell edge probability could possibly result 200 $70 150 in a 1.9 Km cell radius. Assuming all other $94 100 design assumptions are the same the cost per 50 C a p ita l C o s t P e0 r S u b s c rib e r (U S D ) subscriber metric improves tremendously. Base Station Backhaul, Batteries & Construction Costs IP Core Total Netw ork CPE (Blended) Total Cost ( Incl. Allocations) Environmentals (Blended) Per Sub IP EMS/NMS, AAA 75% Indoor 80% Collo/Rooftop DHCP/DNS, SMS 25% PCMCIA 25% Greenfield L3 aggregation Router

Assumptions: 440 HP’s/Km2 market density, 20% penetration, Data & VoIP, 1Mb/s DL & 256 Kb/s UL, 1.32Km cell radius, 2.3 GHz, 3 sector, 10 MHz, N=3 Costs Exclude: OSS/BSS, ISP Costs (email, web, news, chat…), VoIP Platform (Softswitch, MGW…), Interconnect

Chart 9: WiMAX Cost Per Subscriber

An interesting sensitivity to the cost per subscriber number mentioned is the effect market densities and penetration rates have on

WiMAX Costs Per Subscriber For Various Market In conclusion, the economics of WiMAX can be compelling if the network operator has Densities & Subscriber Penetration Levels the wherewithal to spend a large sum of Market E Europe Country US Metro Asia Suburbs W Europe City upfront capital to build a very dense network. Area (KM2) 20,273 12,949 443 830 This expenditure is only justifiable if there is a Homes Passed 509,096 983,000 186,000 425,000 fairly compact addressable market and it is HP/KM2 25 76 420 512 underserved. The definition of underserved s

e must mean that downlink speeds of 1-3 Mb/s t 1.0% $145,225 $48,558 $9,892 $8,281 are not readily available from competitors and Ra

3.0% $48,558 $18,350 $3,447 $2,910

n that uplink speeds of 256 Kb/s are sufficient.

o 10.0% $16,336 $5,595 $875 $758 i

t Additionally, underserved probably means that

a 15.0% $10,582 $3,850 $658 $580 r

t there is a sufficient pent up demand for

e 20.0% $8,281 $2,910 $550 $492

n mobility and portability by customers when 25.0% $6,529 $2,389 $485 $438

Pe using broadband services. And finally, these Assumptions: 1.32 Km cell radius, 3 sector, 10 MHz channel, N=3, 2:1 DL/UL ratio, 1 Mb/s DL & 256 Kb/s UL, Fixed Indoor CPE needs outweigh the inconsistent service levels Table 11: Cost per Sub for Various Market a subscriber will receive because of the Densities inherent variability of wireless technology.

The business case will now start to look Comparison of WiMAX to Fixed Broadband beneficial at low penetration rates (say 10%) in the more densely populated areas. As in Contrasting WiMAX capacities to the well any sensitivity analysis there are a lot of known Cable High Speed Data (HSD) variables to change and WiMAX certainly capabilities of DOCSIS is a very useful does not have its shortage of moving parts. exercise. Table 12 compares the capacities and economics of a DOCSIS 1.x network with All the previous economic analysis two different WiMAX configurations. It assumed a network designed to serve an should be noted that although DOCSIS 1.x indoor high power/high gain type CPE. The was chosen most operators have deployed costs to serve this CPE design result in $325 DOCSIS 2.0 and DOCSIS 3.0 installations are per subscriber of network costs and $225/sub expected to start in 2008. Both advancements of CPE costs. The vision of WiMAX is for will provide even higher capacities to the customers to access a broadband network with Cable side of this comparison. a laptop containing an embedded WiMAX A 10 MHz WiMAX channel bandwidth chipset and software. The economics of a operating at a 3:1 DL/UL ratio with either an laptop CPE network that obtains 1 Mb/s N=1 and N=3 frequency reuse are used. A downlink and 256 Kb/s uplink will cause CPE common customer product speed offering of 1 costs to go down to approximately $100/sub. Mb/s downstream and 256 Kb/s upstream is Unfortunately the associated network costs assumed. Although these are low speeds for a will rise to the $500/sub level in order to cable network, because it appears to be the support a smaller cell radius. Therefore, in “sweet spot” for WiMAX networks, we will total, the costs will go up and no savings are focus first on this scenario. obtained.

Comparison of WiMAX to Cable for 1 Mb/s Downlink and 256 Kb/s Uplink WiMAX Cable Metric 10 MHz, N=1, 10 MHz, N=3, DOCSIS 1.x 3:1 DL/UL, 3:1 DL/UL, Average Throughput DL 8.5 15.0 38.01 (Mbps) UL 1.1 2.2 16.42 Max Capacity (Subs/site or node)4,5 151 378 1,024

Cost/Subscriber6 $ 1,185 $ 608 $ 31 Cost/Mbps DL $ 5,6865 $ 3,222 $ 284 (sector/node 5 capacity) UL $ 43,155 $ 21,970 $ 659 1. DOCSIS 1.x downstream: 256 QAM for a 6 MHz channel bandwidth = 42.88 Mb/s less O/H = 38 Mb/s 2. DOCSIS 1.x Upstream at 16QAM for a 3.2 MHz channel is 10.24 MB/s less O/H (20%) = 8.2 Mb/s, (2) US channels 3. Market Density is 420 HP per Km2 @ 1.32 Km cell radius = 2,300 HP/site. Cable market is 4 nodes @ 575 HP/Node = 2,300 HP/DOCSIS channel 4. WiMAX site costs are $145K and $225 per indoor CPE.. Cable DOCSIS costs are loaded QAM costs of $10,800 per stream and $20 CM/EMTA. Table 12: Comparison of WiMAX to Cable

As the numbers indicate in Table 12 the Looking at the higher capacity customer DOCSIS network has 6.5 and 2.5 times the proposition of an 8 Mb/s downstream and 1 subscriber capacities of the respective Mb/s upstream product in Table 13 clearly WiMAX networks. These results used a indicates the capacity and cost advantage of common HSD capacity traffic model that DOCSIS over wireless broadband. assumed a DS overbooking of 40:1 resulting in an average speed of 25 Kb/s and an US DOCSIS 1.x with two 3.2 MHz upstream overbooking of 25:1 leading to a 10.2 Kb/s channels allocated results in a 3.5 times average speed. The WiMAX capacity improvement in capacity over WiMAX. In the numbers appear slightly different from WiMAX case the upstream is clearly the previous discussions in this paper because a limiting item on the number of subscribers more realistic 30% P2P traffic assumption and serviceable. Furthermore, the cost per equivalent circuit theory tier deduction was subscriber advantage of DOCSIS is amplified accounted for in the calculations. as a $66/sub cost is associated with the Cable network while $2,456/sub is needed to build a The comparative economics illustrate a WiMAX network. The DOCSIS capacity similar DOCSIS advantage of $31/sub versus advantage for serving higher product speeds is $608/sub for WiMAX. In addition to the very apparent in this example. If a cost per larger capacity advantage for DOCSIS, the Mb/s metric is calculated and compared as lower cost for a QAM leads to the much lower shown in the first chart the WiMAX network cable cost per subscriber number. Assuming disadvantage for both the downstream and WiMAX is a new entrant, a fully loaded three upstream is clearly identified. The upstream sector WiMAX site is compared with an economic limitations of WiMAX becomes embedded Cable plant. A more detailed even more pronounced when the $22,000 comparison will want to delve into the impact WiMAX metric is contrasted to the $659 cost of adding cable upgrade costs or even a new per Mb/s DOCSIS number. build into the analysis.

Comparison of WiMAX to Cable for 8 Mb/s Downlink and 1 Mb/s Uplink WiMAX Cable Metric 10 MHz, N=3, DOCSIS 1.x 3:1 DL/UL, Average Throughput DL 15.0 38.01 (Mbps) UL 2.2 16.42 Max Capacity (Subs/site or node)4,5 65 233

Cost/Subscriber6 $ 2,456 $ 66

1. DOCSIS 1.x downstream: 256 QAM for a 6 MHz channel bandwidth = 42.88 Mb/s less O/H = 38 Mb/s 2. DOCSIS 1.x Upstream at 16QAM for a 3.2 MHz channel is 10.24 MB/s less O/H (20%) = 8.2 Mb/s, (2) US channels = 16.4 Mb/s 3. Overbooking; A 1 Mb/s DL product speed uses a 40:1 overbooking which results in an average DL speed of 25 Kb/s. A 256 Kb/s UL product speed uses a 25:1 overbooking which results in an average UL speed of 10.2 Kb/s Table 13: Comparison of WiMAX to Cable

Overall, it is clear from this analysis that a Overall, as a competitor to fixed line fixed broadband solution such as DOCSIS has services, like Cable’s HSD offerings, WiMAX the ability to more easily scale US and DS does not have the speeds or economics to capacity to meet the growing capacity needs of compete against an in place fixed line customers. In fairness to WiMAX capabilities, broadband technology. The uplink challenges, the costs and capacities shown here do offer a variability of service levels to customers and nomadic and portable capability not possible lack of downlink speeds for video will with fixed networks. constrain its competitiveness against fixed line alternatives, Conclusions Once the entire WiMAX ecosystem is built The WiMAX standard extends the state of out and technology, such as MIMO, advances the art for wireless technology. It lays the it may offer a wireless displacement option for foundation for a true 4th generation wireless certain customers. If, however, speeds and technology that offers both a broadband and consumption continue to grow, as they have mobility customer proposition. In any new historically, wireless may have a difficult time technology the advantages are relative to what meeting the home requirements of consumers. it is being compared to. As a comparison to existing wireless technology, WiMAX will Finally, for fixed line operators, the use of exceed the capacity, throughput and economic WiMAX could serve as an extremely capabilities of its CDMA predecessors while complementary lower speed tier that augments still offering full portability and mobility. the high speed fixed capabilities currently available. The ability of an MSO to provide It is the opinion of the authors that the customers with a portable broadband service technical and economic viability of WiMAX that also provides mobility can be a very versus fixed line alternatives like DOCSIS is compelling proposition. very targeted. WiMAX seems to make sense in environments where there is a lack of viable REFERENCES competitor offerings of customer downlink speeds greater than 3 Mb/s and uplink speeds 1. WiMAX Forum web site, “Technical better than 256 Kb/s. Additionally, minimum Overview and Performance Evaluation”, market density environments and realistic March 2006 http://wimaxforum.org target penetration rates must be attainable for the economics to work.

2. WiMAX Forum web site, “The Business 17. Nortel Technical Journal, “Wireless Case for Fixed Access in Emerging Access”, July, 2006 Markets”, March 2005 18. Intel Technical Journal, “SOFDMA in http://wimaxforum.org 802.16e”, Vol 8, Issue 3, 2004 3. WiMAX Forum web site, “WiMAX End to 19. Intel, “The Business Case for WiMAX”, End Network System Architecture”, April February 2007 2005 http://wimaxforum.org 20. Intel Technical Journal, “Multiple Antenna 4. WiMAX Forum web site, “Considerations Technology in WiMAX Systems”, Issue 3, for Fixed Wireless in 2.5, 3.5 GHz”, June, 2004 2005 http://wimaxforum.org 21. HSBC Global WiMAX, “WiMAX Anti- 5. WiMAX Forum web site, “Mobile WiMAX a Climax”, Richard Dineen, January 3, 2006 Comparative Analysis”, March 2006 http://wimaxforum.org 22. Rethink Research Associates, “Wireless Watch” “February 19, 2007 6. Jeffrey Andrews, "Orthogonal Frequency Division Multiple Access" IEEE Journal, 23. Rethink Research Associates, “WiMAX July 29, 2006 Business Models; How to Make Money in WiMAX” “October 17, 2005 7. Diversity, Interference& Spatial Multiplexing in MIMO Mobile WiMAX”, E. 24. Marvedis, “WiMAX and Broadband Biglieri, 2007 Worldwide Market Trends and Analysis” “September, 2006 8. IEEE Standard 802.16e – 2005 and Amendment & Corrigendum to 802.16- 25. Navini White Paper, “VoIP over WiMAX” 2004, IEEEE LAN/MAN Standards “November, 2006 Committee, February 28, 2006 26. SRTelecom White Paper, “WiMAX 9. Alcatel Technology White Paper, Capacity” “November, 2006 “Universal WiMAX”, September14, 2006 27. Ericsson White Paper, “Basic Concepts of 10. Alcatel Technology White Paper, HSDPA” February, 2007 “Extending the DSL Penetration to Mobile 28. Qualcomm White Paper, “The Economics of WiMAX”, 2Q, 2005 Mobile Wireless Data” August, 2006 11. Alcatel Technology White Paper, “4G”, 29. Qualcomm White Paper, “A Wireless September 14, 2006 Primer”, Jeff Belk, September, 2005 12. Alcatel Technology White Paper, “WiMAX, 30. Alvarion White Paper, “ Understanding the Making Ubquitous High Speed Data a Radio Technology of Mobile WiMAX” Reality”, June 28, 2004 2006 13. Alcatel Technology White Paper, 31. Alvarion White Paper, “ Comparing Mobile “Universal Broadband Access Going WiMAX 3G and Beyond” 2006 Wireless and Mobile”, 2Q05 32. CableLabs, “Cable Optimized Mobile 14. Nortel White Paper, “MIMO, Key Wireless Broadband”, November 4, 2005 Technology Choice in Deploying WiMAX”, October 10, 2006 33. CableLabs, “Cost Comparison of Mobile Wireless Technologies”, June, 2005 15. Nortel WiMAX Solution Brief, March, 2007

16. Nortel White Paper, “Considerations for Deploying WiMAX at various Frequencies”, March, 2006

ISBN 0‐940272‐01‐6; 0‐940272‐08‐3; 0‐940272‐10‐5; 0‐940272‐11‐3; 0‐940272‐12‐1; 0‐940272‐14‐8; 0‐ 940272‐15‐6; 0‐940272‐16‐4; 0‐940272‐18‐0; 0‐940272‐19‐9; 0‐940272‐20‐2; 0‐940272‐21‐0; 0‐940272‐ 22‐9; 0‐940272‐23‐7; 0‐940272‐24‐5; 0‐940272‐25‐3; 0‐940272‐26‐1; 0‐940272‐27‐X; 0‐940272‐28‐8; 0‐ 940272‐29‐6; 0‐940272‐32‐6; 0‐940272‐33‐4; 0‐940272‐34‐2; 0‐940272‐35‐0; 0‐940272‐36‐9; 0‐940272‐ 37‐7; 0‐940272‐38‐5; 0‐940272‐39‐3; 0‐940272‐40‐7; 0‐940272‐41‐5; 0‐940272‐42‐3; 0‐940272‐43‐1; 0‐ 940272‐44‐X; 0‐940272‐45‐8; 0‐940272‐46‐6; 0‐940272‐47‐4; 0‐940272‐48‐2; 0‐940272‐49‐0; 0‐940272‐ 50‐4; 0‐940272‐51‐2; 0‐940272‐52‐0; 0‐940272‐53‐9; 0‐940272‐54‐7

© 2015 National Cable and Telecommunications Association. All Rights Reserved.